You are on page 1of 551

MICROWAVE DEVICES,

CIRCUITS AND
SUBSYSTEMS FOR
COMMUNICATIONS
ENGINEERING
Edited by

I. A. Glover, S. R. Pennock and P. R. Shepherd


All of
Department of Electronic and Electrical Engineering
University of Bath, UK
MICROWAVE DEVICES,
CIRCUITS AND
SUBSYSTEMS FOR
COMMUNICATIONS
ENGINEERING
MICROWAVE DEVICES,
CIRCUITS AND
SUBSYSTEMS FOR
COMMUNICATIONS
ENGINEERING
Edited by

I. A. Glover, S. R. Pennock and P. R. Shepherd


All of
Department of Electronic and Electrical Engineering
University of Bath, UK
Copyright © 2005 John Wiley & Sons Ltd, The Atrium, Southern Gate, Chichester,
West Sussex PO19 8SQ, England

Telephone (+44) 1243 779777

Email (for orders and customer service enquiries): cs-books@wiley.co.uk


Visit our Home Page on www.wiley.com

All Rights Reserved. No part of this publication may be reproduced, stored in a retrieval system or
transmitted in any form or by any means, electronic, mechanical, photocopying, recording, scanning or
otherwise, except under the terms of the Copyright, Designs and Patents Act 1988 or under the terms of a
licence issued by the Copyright Licensing Agency Ltd, 90 Tottenham Court Road, London W1T 4LP, UK,
without the permission in writing of the Publisher. Requests to the Publisher should be addressed to the
Permissions Department, John Wiley & Sons Ltd, The Atrium, Southern Gate, Chichester, West Sussex PO19
8SQ, England, or emailed to permreq@wiley.co.uk, or faxed to (+44) 1243 770620.

Designations used by companies to distinguish their products are often claimed as trademarks. All brand names
and product names used in this book are trade names, service marks, trademarks or registered trademarks of
their respective owners. The Publisher is not associated with any product or vendor mentioned in this book.

This publication is designed to provide accurate and authoritative information in regard to the subject matter
covered. It is sold on the understanding that the Publisher is not engaged in rendering professional services. If
professional advice or other expert assistance is required, the services of a competent professional should be
sought.

Other Wiley Editorial Offices

John Wiley & Sons Inc., 111 River Street, Hoboken, NJ 07030, USA

Jossey-Bass, 989 Market Street, San Francisco, CA 94103-1741, USA

Wiley-VCH Verlag GmbH, Boschstr. 12, D-69469 Weinheim, Germany

John Wiley & Sons Australia Ltd, 33 Park Road, Milton, Queensland 4064, Australia

John Wiley & Sons (Asia) Pte Ltd, 2 Clementi Loop #02-01, Jin Xing Distripark, Singapore 129809

John Wiley & Sons Canada Ltd, 22 Worcester Road, Etobicoke, Ontario, Canada M9W 1L1

Wiley also publishes its books in a variety of electronic formats. Some content that appears
in print may not be available in electronic books.

British Library Cataloguing in Publication Data

A catalogue record for this book is available from the British Library

ISBN 0-471-89964-X (HB)

Typeset in 10/12pt Times by Graphicraft Limited, Hong Kong, China.


Printed and bound in Great Britain by Antony Rowe Ltd, Chippenham, Wiltshire.
This book is printed on acid-free paper responsibly manufactured from sustainable forestry
in which at least two trees are planted for each one used for paper production.
Contents v

Contents

List of Contributors xv

Preface xvii

1 Overview 1
I. A. Glover, S. R. Pennock and P. R. Shepherd
1.1 Introduction 1
1.2 RF Devices 2
1.3 Signal Transmission and Network Methods 4
1.4 Amplifiers 5
1.5 Mixers 6
1.6 Filters 7
1.7 Oscillators and Frequency Synthesisers 7

2 RF Devices: Characteristics and Modelling 9


A. Suarez and T. Fernandez
2.1 Introduction 9
2.2 Semiconductor Properties 10
2.2.1 Intrinsic Semiconductors 10
2.2.2 Doped Semiconductors 13
2.2.2.1 N-type doping 13
2.2.2.2 P-type doping 14
2.2.3 Band Model for Semiconductors 14
2.2.4 Carrier Continuity Equation 17
2.3 P-N Junction 18
2.3.1 Thermal Equilibrium 18
2.3.2 Reverse Bias 21
2.3.3 Forward Bias 23
2.3.4 Diode Model 24
2.3.5 Manufacturing 25
2.3.6 Applications of P-N Diodes at Microwave Frequencies 26
2.3.6.1 Amplitude modulators 28
2.3.6.2 Phase shifters 29
2.3.6.3 Frequency multipliers 30
2.4 The Schottky Diode 32
2.4.1 Thermal Equilibrium 32
2.4.2 Reverse Bias 34
vi Contents

2.4.3 Forward Bias 35


2.4.4 Electric Model 36
2.4.5 Manufacturing 37
2.4.6 Applications 37
2.4.6.1 Detectors 38
2.4.6.2 Mixers 39
2.5 PIN Diodes 40
2.5.1 Thermal Equilibrium 40
2.5.2 Reverse Bias 40
2.5.3 Forward Bias 41
2.5.4 Equivalent Circuit 43
2.5.5 Manufacturing 44
2.5.6 Applications 45
2.5.6.1 Switching 45
2.5.6.2 Phase shifting 47
2.5.6.3 Variable attenuation 50
2.5.6.4 Power limiting 50
2.6 Step-Recovery Diodes 51
2.7 Gunn Diodes 52
2.7.1 Self-Oscillations 54
2.7.2 Operating Modes 55
2.7.2.1 Accumulation layer mode 56
2.7.2.2 Transit-time dipole layer mode 56
2.7.2.3 Quenched dipole layer mode 56
2.7.2.4 Limited-space-charge accumulation (LSA) mode 57
2.7.3 Equivalent Circuit 57
2.7.4 Applications 58
2.7.4.1 Negative resistance amplifiers 58
2.7.4.2 Oscillators 59
2.8 IMPATT Diodes 59
2.8.1 Doping Profiles 60
2.8.2 Principle of Operation 60
2.8.3 Device Equations 62
2.8.4 Equivalent Circuit 63
2.9 Transistors 65
2.9.1 Some Preliminary Comments on Transistor Modelling 65
2.9.1.1 Model types 65
2.9.1.2 Small and large signal behaviour 65
2.9.2 GaAs MESFETs 66
2.9.2.1 Current-voltage characteristics 68
2.9.2.2 Capacitance-voltage characteristics 70
2.9.2.3 Small signal equivalent circuit 71
2.9.2.4 Large signal equivalent circuit 74
2.9.2.5 Curtice model 74
2.9.3 HEMTs 75
2.9.3.1 Current-voltage characteristics 76
2.9.3.2 Capacitance-voltage characteristics 78
2.9.3.3 Small signal equivalent circuit 78
2.9.3.4 Large signal equivalent circuit 78
Contents vii

2.9.4 HBTs 80
2.9.4.1 Current-voltage characteristics 84
2.9.4.2 Capacitance-voltage characteristics 84
2.9.4.3 Small signal equivalent circuit 86
2.9.4.4 Large signal equivalent circuit 87
2.10 Problems 88
References 89

3 Signal Transmission, Network Methods and Impedance Matching 91


N. J. McEwan, T. C. Edwards, D. Dernikas and I. A. Glover
3.1 Introduction 91
3.2 Transmission Lines: General Considerations 92
3.2.1 Structural Classification 92
3.2.2 Mode Classes 94
3.3 The Two-Conductor Transmission Line: Revision of
Distributed Circuit Theory 95
3.3.1 The Differential Equations and Wave Solutions 96
3.3.2 Characteristic Impedance 98
3.4 Loss, Dispersion, Phase and Group Velocity 99
3.4.1 Phase Velocity 100
3.4.2 Loss 100
3.4.3 Dispersion 101
3.4.4 Group Velocity 102
3.4.5 Frequency Dependence of Line Parameters 105
3.4.5.1 Frequency dependence of G 108
3.4.6 High Frequency Operation 109
3.4.6.1 Lossless approximation 111
3.4.6.2 The telegrapher’s equation and the wave equation 111
3.5 Field Theory Method for Ideal TEM Case 113
3.5.1 Principles of Electromagnetism: Revision 114
3.5.2 The TEM Line 117
3.5.3 The Static Solutions 117
3.5.4 Validity of the Time Varying Solution 119
3.5.5 Features of the TEM Mode 121
3.5.5.1 A useful relationship 122
3.5.6 Picturing the Wave Physically 123
3.6 Microstrip 126
3.6.1 Quasi-TEM Mode and Quasi-Static Parameters 128
3.6.1.1 Fields and static TEM design parameters 128
3.6.1.2 Design aims 129
3.6.1.3 Calculation of microstrip physical width 130
3.6.2 Dispersion and its Accommodation in Design Approaches 132
3.6.3 Frequency Limitations: Surface Waves and Transverse Resonance 135
3.6.4 Loss Mechanisms 137
3.6.5 Discontinuity Models 139
3.6.5.1 The foreshortened open end 139
3.6.5.2 Microstrip vias 141
3.6.5.3 Mitred bends 142
3.6.5.4 The microstrip T-junction 142
viii Contents

3.6.6 Introduction to Filter Construction Using Microstrip 145


3.6.6.1 Microstrip low-pass filters 145
3.6.6.2 Example of low-pass filter design 148
3.7 Coupled Microstrip Lines 148
3.7.1 Theory Using Even and Odd Modes 150
3.7.1.1 Determination of coupled region physical length 156
3.7.1.2 Frequency response of the coupled region 157
3.7.1.3 Coupler directivity 158
3.7.1.4 Coupler compensation by means of lumped capacitors 159
3.7.2 Special Couplers: Lange Couplers, Hybrids and Branch-Line
Directional Couplers 161
3.8 Network Methods 163
3.8.1 Revision of z, y, h and ABCD Matrices 164
3.8.2 Definition of Scattering Parameters 166
3.8.3 S-Parameters for One- and Two-Port Networks 168
3.8.4 Advantages of S-Parameters 171
3.8.5 Conversion of S-Parameters into Z-Parameters 171
3.8.6 Non-Equal Complex Source and Load Impedance 174
3.9 Impedance Matching 176
3.9.1 The Smith Chart 176
3.9.2 Matching Using the Smith Chart 182
3.9.2.1 Lumped element matching 182
3.9.2.2 Distributed element matching 187
3.9.2.3 Single stub matching 187
3.9.2.4 Double stub matching 189
3.9.3 Introduction to Broadband Matching 191
3.9.4 Matching Using the Quarter Wavelength Line Transformer 194
3.9.5 Matching Using the Single Section Transformer 194
3.10 Network Analysers 195
3.10.1 Principle of Operation 196
3.10.1.1 The signal source 197
3.10.1.2 The two-port test set 197
3.10.1.3 The receiver 198
3.10.2 Calibration Kits and Principles of Error Correction 198
3.10.3 Transistor Mountings 202
3.10.4 Calibration Approaches 206
3.11 Summary 207
References 208

4 Amplifier Design 209


N. J. McEwan and D. Dernikas
4.1 Introduction 209
4.2 Amplifier Gain Definitions 209
4.2.1 The Transducer Gain 211
4.2.2 The Available Power Gain 212
4.2.3 The Operating Power Gain 213
4.2.4 Is There a Fourth Definition? 213
4.2.5 The Maximum Power Transfer Theorem 213
4.2.6 Effect of Load on Input Impedance 216
4.2.7 The Expression for Transducer Gain 218
Contents ix

4.2.8 The Origin of Circle Mappings 221


4.2.9 Gain Circles 222
4.3 Stability 223
4.3.1 Oscillation Conditions 224
4.3.2 Production of Negative Resistance 227
4.3.3 Conditional and Unconditional Stability 228
4.3.4 Stability Circles 229
4.3.5 Numerical Tests for Stability 230
4.3.6 Gain Circles and Further Gain Definitions 231
4.3.7 Design Strategies 237
4.4 Broadband Amplifier Design 239
4.4.1 Compensated Matching Example 240
4.4.2 Fano’s Limits 241
4.4.3 Negative Feedback 243
4.4.4 Balanced Amplifiers 244
4.4.4.1 Principle of operation 245
4.4.4.2 Comments 245
4.4.4.3 Balanced amplifier advantages 246
4.4.4.4 Balanced amplifier disadvantages 246
4.5 Low Noise Amplifier Design 246
4.5.1 Revision of Thermal Noise 246
4.5.2 Noise Temperature and Noise Figure 248
4.5.3 Two-Port Noise as a Four Parameter System 250
4.5.4 The Dependence on Source Impedance 251
4.5.5 Noise Figures Circles 254
4.5.6 Minimum Noise Design 255
4.6 Practical Circuit Considerations 256
4.6.1 High Frequencies Components 256
4.6.1.1 Resistors 256
4.6.1.2 Capacitors 259
4.6.1.3 Capacitor types 261
4.6.1.4 Inductors 263
4.6.2 Small Signal Amplifier Design 267
4.6.2.1 Low-noise amplifier design using CAD software 268
4.6.2.2 Example 269
4.6.3 Design of DC Biasing Circuit for Microwave Bipolar Transistors 272
4.6.3.1 Passive biasing circuits 272
4.6.3.2 Active biasing circuits 274
4.6.4 Design of Biasing Circuits for GaAs FET Transistors 277
4.6.4.1 Passive biasing circuits 277
4.6.4.2 Active biasing circuits 279
4.6.5 Introduction of the Biasing Circuit 279
4.6.5.1 Implementation of the RFC in the bias network 282
4.6.5.2 Low frequency stability 287
4.6.5.3 Source grounding techniques 288
4.7 Computer Aided Design (CAD) 290
4.7.1 The RF CAD Approach 291
4.7.2 Modelling 293
4.7.3 Analysis 296
4.7.3.1 Linear frequency domain analysis 296
x Contents

4.7.3.2 Non-linear time domain transient analysis 297


4.7.3.3 Non-linear convolution analysis 297
4.7.3.4 Harmonic balance analysis 297
4.7.3.5 Electromagnetic analysis 298
4.7.3.6 Planar electromagnetic simulation 298
4.7.4 Optimisation 298
4.7.4.1 Optimisation search methods 299
4.7.4.2 Error function formulation 300
4.7.5 Further Features of RF CAD Tools 302
4.7.5.1 Schematic capture of circuits 302
4.7.5.2 Layout-based design 302
4.7.5.3 Statistical design of RF circuits 303
Appendix I 306
Appendix II 306
References 310

5 Mixers: Theory and Design 311


L. de la Fuente and A. Tazon
5.1 Introduction 311
5.2 General Properties 311
5.3 Devices for Mixers 313
5.3.1 The Schottky-Barrier Diode 313
5.3.1.1 Non-linear equivalent circuit 313
5.3.1.2 Linear equivalent circuit at an operating point 314
5.3.1.3 Experimental characterization of Schottky diodes 317
5.3.2 Bipolar Transistors 319
5.3.3 Field-Effect Transistors 321
5.4 Non-Linear Analysis 322
5.4.1 Intermodulation Products 323
5.4.2 Application to the Schottky-Barrier Diode 327
5.4.3 Intermodulation Power 327
5.4.4 Linear Approximation 329
5.5 Diode Mixer Theory 331
5.5.1 Linear Analysis: Conversion Matrices 332
5.5.1.1 Conversion matrix of a non-linear resistance/conductance 333
5.5.1.2 Conversion matrix of a non-linear capacitance 335
5.5.1.3 Conversion matrix of a linear resistance 336
5.5.1.4 Conversion matrix of the complete diode 337
5.5.1.5 Conversion matrix of a mixer circuit 337
5.5.1.6 Conversion gain and input/output impedances 338
5.5.2 Large Signal Analysis: Harmonic Balance Simulation 339
5.6 FET Mixers 341
5.6.1 Single-Ended FET Mixers 341
5.6.1.1 Simplified analysis of a single-gate FET mixer 341
5.6.1.2 Large-signal and small-signal analysis of single-gate FET mixers 343
5.6.1.3 Other topologies 346
5.7 Double–Gate FET Mixers 349
5.7.1 IF Amplifier 354
5.7.2 Final Design 355
5.7.3 Mixer Measurements 356
Contents xi

5.8 Single-Balanced FET Mixers 358


5.9 Double-Balanced FET Mixers 359
5.10 Harmonic Mixers 360
5.10.1 Single-Device Harmonic Mixers 362
5.10.2 Balanced Harmonic Mixers 362
5.11 Monolithic Mixers 364
5.11.1 Characteristics of Monolithic Medium 365
5.11.2 Devices 366
5.11.3 Single-Device FET Mixers 366
5.11.4 Single-Balanced FET Mixers 368
5.11.5 Double-Balanced FET Mixers 370
Appendix I 375
Appendix II 375
References 376

6 Filters 379
A. Mediavilla
6.1 Introduction 379
6.2 Filter Fundamentals 379
6.2.1 Two-Port Network Definitions 379
6.2.2 Filter Description 381
6.2.3 Filter Implementation 383
6.2.4 The Low Pass Prototype Filter 383
6.2.5 The Filter Design Process 384
6.2.5.1 Filter simulation 384
6.3 Mathematical Filter Responses 385
6.3.1 The Butterworth Response 385
6.3.2 The Chebyshev Response 386
6.3.3 The Bessel Response 390
6.3.4 The Elliptic Response 390
6.4 Low Pass Prototype Filter Design 393
6.4.1 Calculations for Butterworth Prototype Elements 395
6.4.2 Calculations for Chebyshev Prototype Elements 400
6.4.3 Calculations for Bessel Prototype Elements 404
6.4.4 Calculations for Elliptic Prototype Elements 405
6.5 Filter Impedance and Frequency Scaling 405
6.5.1 Impedance Scaling 405
6.5.2 Frequency Scaling 410
6.5.3 Low Pass to Low Pass Expansion 410
6.5.4 Low Pass to High Pass Transformation 412
6.5.5 Low Pass to Band Pass Transformation 414
6.5.6 Low Pass to Band Stop Transformation 418
6.5.7 Resonant Network Transformations 421
6.6 Elliptic Filter Transformation 423
6.6.1 Low Pass Elliptic Translation 423
6.6.2 High Pass Elliptic Translation 425
6.6.3 Band Pass Elliptic Translation 426
6.6.4 Band Stop Elliptic Translation 426
6.7 Filter Normalisation 429
6.7.1 Low Pass Normalisation 429
xii Contents

6.7.2 High Pass Normalisation 430


6.7.3 Band Pass Normalisation 431
6.7.3.1 Broadband band pass normalisation 432
6.7.3.2 Narrowband band pass normalisation 433
6.7.4 Band Stop Normalisation 435
6.7.4.1 Broadband band stop normalisation 436
6.7.7.2 Narrowband band stop normalisation 438

7 Oscillators, Frequency Synthesisers and PLL Techniques 461


E. Artal, J. P. Pascual and J. Portilla
7.1 Introduction 461
7.2 Solid State Microwave Oscillators 461
7.2.1 Fundamentals 461
7.2.1.1 An IMPATT oscillator 463
7.2.2 Stability of Oscillations 466
7.3 Negative Resistance Diode Oscillators 467
7.3.1 Design Technique Examples 469
7.4 Transistor Oscillators 469
7.4.1 Design Fundamentals of Transistor Oscillators 471
7.4.1.1 Achievement of the negative resistance 472
7.4.1.2 Resonator circuits for transistor oscillators 473
7.4.2 Common Topologies of Transistor Oscillators 475
7.4.2.1 The Colpitts oscillator 476
7.4.2.2 The Clapp oscillator 477
7.4.2.3 The Hartley oscillator 477
7.4.2.4 Other practical topologies of transistor oscillators 478
7.4.2.5 Microwave oscillators using distributed elements 478
7.4.3 Advanced CAD Techniques of Transistor Oscillators 479
7.5 Voltage-Controlled Oscillators 481
7.5.1 Design Fundamentals of Varactor-Tuned Oscillators 481
7.5.2 Some Topologies of Varactor-Tuned Oscillators 482
7.5.2.1 VCO based on the Colpitts topology 482
7.5.2.2 VCO based on the Clapp topology 483
7.5.2.3 Examples of practical topologies of microwave VCOs 483
7.6 Oscillator Characterisation and Testing 484
7.6.1 Frequency 485
7.6.2 Output Power 485
7.6.3 Stability and Noise 485
7.6.3.1 AM and PM noise 486
7.6.4 Pulling and Pushing 488
7.7 Microwave Phase Locked Oscillators 489
7.7.1 PLL Fundamentals 489
7.7.2 PLL Stability 493
7.8 Subsystems for Microwave Phase Locked Oscillators (PLOs) 493
7.8.1 Phase Detectors 494
7.8.1.1 Exclusive-OR gate 495
7.8.1.2 Phase-frequency detectors 496
7.8.2 Loop Filters 501
7.8.3 Mixers and Harmonic Mixers 505
7.8.4 Frequency Multipliers and Dividers 506
Contents xiii

7.8.4.1 Dual modulus divider 506


7.8.4.2 Multipliers 508
7.8.5 Synthesiser ICs 508
7.9 Phase Noise 509
7.9.1 Free running and PLO Noise 513
7.9.1.1 Effect of multiplication in phase noise 514
7.9.2 Measuring Phase Noise 514
7.10 Examples of PLOs 514
References 518

Index 519
xiv Contents
List of Contributors

E. Artal ETSIIT – DICOM, Av. Los Castros, 39005, Santander, Cantabria, Spain.
L. de la Fuente ETSIIT – DICOM, Av. Los Castros, 39005, Santander, Cantabria, Spain.
D. Dernikas Formerly University of Bradford, U.K. Currently Aircom International,
Grosvenor House, 65–71 London Road, Redhill, Surrey, RH1 1LQ,
U.K.
T. C. Edwards Engalco, 3, Georgian Mews, Bridlington, East Yorkshire, YO15 3TG,
U.K.
T. Fernandez ETSIIT – DICOM, Av. Los Castros, 39005, Santander, Cantabria, Spain.
I. A. Glover Department of Electronic and Electrical Engineering, University of
Bath, Claverton Down, Bath, BA2 7AY, U.K.
N. J. McEwan Filtronic PLC, The Waterfront, Salts Mill Road, Saltaire, Shipley, West
Yorkshire, BD18 3TT, U.K.
A. Mediavilla ETSIIT – DICOM, Av. Los Castros, 39005, Santander, Cantabria, Spain.
J. P. Pascual ETSIIT – DICOM, Av. Los Castros, 39005, Santander, Cantabria, Spain.
S. R. Pennock Department of Electronic and Electrical Engineering, University of
Bath, Claverton Down, Bath, BA2 7AY, U.K.
J. Portilla ETSIIT – DICOM, Av. Los Castros, 39005, Santander, Cantabria, Spain.
P. R. Shepherd Department of Electronic and Electrical Engineering, University of
Bath, Claverton Down, Bath, BA2 7AY, U.K.
A. Suarez ETSIIT – DICOM, Av. Los Castros, 39005, Santander, Cantabria, Spain.
A. Tazon ETSIIT – DICOM, Av. Los Castros, 39005, Santander, Cantabria, Spain.
Preface xvii

Preface

This text originated from a Master’s degree in RF Communications Engineering offered


since the mid-1980s at the University of Bradford in the UK. The (one-year) degree, which
has now graduated several hundred students, was divided into essentially three parts:

Part 1 – RF devices and subsystems


Part 2 – RF communications systems
Part 3 – Dissertation project.

Part 1 was delivered principally in Semester 1 (October to mid-February), Part 2 in


Semester 2 (mid-February to June) and Part 3 during the undergraduate summer vacation
(July to September). Parts 1 and 2 comprised the taught component of the degree consisting
of lectures, tutorials, laboratory work and design exercises. Part 3 comprised an indi-
vidual and substantial project drawing on skills acquired in Parts 1 and 2 for its successful
completion.
In the mid-1990s it was decided that a distance-learning version of the degree should
be offered which would allow practising scientists and technologists to retrain as RF and
microwave communications engineers. (At that time there was a European shortage of such
engineers and the perception was that a significant market existed for the conversion of
numerate graduates from other disciplines, e.g. physics and maths, and the retraining of
existing engineers from other specialisations, e.g. digital electronics and software design.)
In order to broaden the market yet further, it was intended that the University of Bradford
would collaborate with other European universities running similar degree programmes
so that the text could be expanded for use in all. The final list of collaborating institu-
tions was:

University of Bradford, UK
University of Cantabria, Spain
University of Bologna, Italy
Telecommunications Systems Institute/Technical University of Crete, Greece

Microwave Devices, Circuits and Subsystems for Communications Engineering is a result of


this collaboration and contains the material delivered in Part 1 of the Bradford degree plus
additional material required to match courses delivered at the other institutions.
xviii Preface

In addition to benefiting students studying the relevant degrees in the collaborating insti-
tutions, it is hoped that the book will prove useful to both the wider student population and
to the practising engineer looking for a refresher or conversion text.
A companion website containing a sample chapter, solutions to selected problems and
figures in electronic form (for the use of instructors adopting the book as a course text) is
available at ftp://ftp.wiley.co.uk/pub/books/glover.
Overview 1

1
Overview
I. A. Glover, S. R. Pennock and P. R. Shepherd

1.1 Introduction
RF and microwave engineering has innumerable applications, from radar (e.g. for air traffic
control and meteorology) through electro-heat applications (e.g. in paper manufacture and
domestic microwave ovens), to radiometric remote sensing of the environment, continuous
process measurements and non-destructive testing. The focus of the courses for which this
text was written, however, is microwave communications and so, while much of the material
that follows is entirely generic, the selection and presentation of material are conditioned by
this application.
Figure 1.1 shows a block diagram of a typical microwave communications transceiver. The
transmitter comprises an information source, a baseband signal processing unit, a modulator,
some intermediate frequency (IF) filtering and amplification, a stage of up-conversion to
the required radio frequency (RF) followed by further filtering, high power amplification
(HPA) and an antenna. The baseband signal processing typically includes one, more, or
all of the following: an antialising filter, an analogue-to-digital converter (ADC), a source
coder, an encryption unit, an error controller, a multiplexer and a pulse shaper. The antialisaing
filter and ADC are only required if the information source is analogue such as a speech
signal, for example. The modulator impresses the (processed) baseband information onto the
IF carrier. (An IF is used because modulation, filtering and amplification are technologically
more difficult, and therefore more expensive, at the microwave RF.)
The receiver comprises an antenna, a low noise amplifier (LNA), microwave filtering, a
down-converter, IF filtering and amplification, a demodulator/detector and a baseband pro-
cessing unit. The demodulator may be coherent or incoherent. The signal processing will
incorporate demultiplexing, error detection/correction, deciphering, source decoding, digital-
to-analogue conversion (DAC), where appropriate, and audio/video amplification and filtering,
again where appropriate. If detection is coherent, phase locked loops (PLLs) or their equivalent
will feature in the detector design. Other control circuits, e.g. automatic gain control (AGC),
may also be present in the receiver.
The various subsystems of Figure 1.1 (and the devices comprising them whether discrete
or in microwave integrated circuit form) are typically connected together with transmission

Microwave Devices, Circuits and Subsystems for Communications Engineering Edited by I. A. Glover, S. R. Pennock
and P. R. Shepherd
© 2005 John Wiley & Sons, Ltd.
2 Microwave Devices, Circuits and Subsystems for Communications Engineering

Information Signal Modulator BPF


Gain
source Processing

BPF BPF
HPA Gain ×

Up-converter ∼

(a)

BPF BPF
LNA × Gain

∼ Down-converter

Information Signal Demodulator/ BPF


Gain
sink processing decision cct

(b)

Figure 1.1 Typical microwave communications transmitter (a) and receiver (b)

lines implemented using a variety of possible technologies (e.g. coaxial cable, microstrip,
co-planar waveguide).
This text is principally concerned with the operating principles and design of the RF/
microwave subsystems of Figure 1.1, i.e. the amplifiers, filters, mixers, local oscillators and
connecting transmission lines. It starts, however, by reviewing the solid-state devices (diodes,
transistors, etc.) incorporated in most of these subsystems since, assuming good design, it is
the fundamental physics of these devices that typically limits performance.
Sections 1.2–1.7 represent a brief overview of the material in each of the following chapters.

1.2 RF Devices
Chapter 2 begins with a review of semiconductors, their fundamental properties and the
features that distinguish them from conductors and insulators. The role of electrons and
holes as charge carriers in intrinsic (pure) semiconductors is described and the related con-
cepts of carrier mobility, drift velocity and drift current are presented. Carrier concentration
gradients, the diffusion current that results from them and the definition of the diffusion
Overview 3

coefficient are also examined and the doping of semiconductors with impurities to increase
the concentration of electrons or holes is described. A discussion of the semiconductor
energy-band model, which underlies an understanding of semiconductor behaviour, is pre-
sented and the important concept of the Fermi energy level is defined. This introductory
but fundamental review of semiconductor properties finishes with the definition of mean
carrier lifetime and an outline derivation of the carrier continuity equation, which plays a
central role in device physics.
Each of the next six major sections deals with a particular type of semiconductor diode.
In order of treatment these are (i) simple P-N junctions; (ii) Schottky diodes; (iii) PIN
diodes; (iv) step-recovery diodes; (v) Gunn diodes; and (vi) IMPATT diodes. (The use of
the term diode in the context of Gunn devices is questionable but almost universal and
so we choose here to follow convention.) The treatment of the first three diode types
follows the same pattern. The device is first described in thermal equilibrium (i.e. with
no externally applied voltage), then under conditions of reverse bias (the P-material being
made negative with respect to the N-material), and finally under conditions of forward bias
(the P-material being made positive with respect to the N-material). Following discussion
of the device’s physics under these different conditions, an equivalent circuit model is
presented that, to an acceptable engineering approximation, emulates the device’s terminal
behaviour. It is a device’s equivalent circuit model that is used in the design of circuits and
subsystems. There is a strong modern trend towards computer-aided design in which case
the equivalent circuit models (although of perhaps greater sophistication and accuracy than
those presented here) are incorporated in the circuit analysis software. The discussion of
each device ends with some comments about its manufacture and a description of some
typical applications.
The treatment of the following diode types is less uniform. Step-recovery diodes, being a
variation on the basic PIN diode, are described only briefly. The Gunn diode is discussed
in some detail since its operating principles are quite different from those of the previous
devices. Its important negative resistance property, resulting in its principal application in
oscillators and amplifiers, is explained and the relative advantages of its different operating
modes are reviewed. Finally, IMPATT diodes are described, that, like Gunn devices, exhibit
negative resistance and are used in high power (high frequency) amplifiers and oscillators,
their applications being somewhat restricted, however, by their relatively poor noise
characteristics. The doping profiles and operating principles of the IMPATT diode are
described and the important device equations are presented. The discussion of IMPATT
diodes concludes with their equivalent circuit.
Probably the most important solid-state device of all in modern-day electronic engineer-
ing is the transistor and it is this device, in several of its high frequency variations, that is
addressed next. The treatment of transistors starts with some introductory and general
remarks about transistor modelling, in particular, pointing out the difference between small
and large signal models. After these introductory remarks three transistor types are addressed
in turn, all suitable for RF/microwave applications (to a greater or lesser extent). These are
(i) the gallium arsnide metal semiconductor field effect transistor (GaAs MESFET); (ii) the
high electron mobility transistor (HEMT); and (iii) the heterojunction bipolar transistor
(HBT). In each case the treatment is essentially the same: a short description followed by
presentations of the current-voltage characteristic, capacitance-voltage characteristic, the
small signal equivalent circuit and the large signal equivalent circuit.
4 Microwave Devices, Circuits and Subsystems for Communications Engineering

1.3 Signal Transmission and Network Methods


Chapter 3 starts with a survey of practical transmission line structures including those
without conductors (dielectric waveguides), those with a single conductor (e.g. conventional
waveguide), and those with two conductors (e.g. microstrip). With one exception, all the two-
conductor transmission line structures are identified as supporting a quasi-TEM (transverse
electromagnetic) mode of propagation – important because this type of propagation can
be modelled using classical distributed-circuit transmission line theory. A thorough treat-
ment of this theory is given, starting with the fundamental differential equations con-
taining voltage, current and distributed inductance (L), conductance (G), resistance (R) and
capacitance (C), and deriving the resulting line’s attenuation constant, phase constant and
characteristic impedance. Physical interpretations of the solution of the transmission line
equations are given in terms of forward and backward travelling waves and the concepts
of loss, dispersion, group velocity and phase velocity are introduced. The frequency-
dependent behaviour of a transmission line due to the frequency dependence of its L, G, R
and C (due in part to the skin effect) is examined and the special properties of a lossless line
(with R = G = 0) are derived.
Following the distributed-circuit description of transmission lines, the more rigorous field
theory approach to their analysis is outlined. A short revision of fundamental electromagnetic
theory is given before this theory is applied to the simplest (TEM) types of transmission line
with a uniform dielectric and perfect conductors. The relationship between the time-varying
field on the TEM line and the static field solution to Maxwell’s equations is discussed and
the validity of the solutions derived from this relationship is confirmed. The special charac-
teristics of the TEM propagation mode are examined in some detail. The discussion of basic
transmission line theory ends with a physical interpretation of the field solutions and a
visualisation of the field distribution in a coaxial line.
Most traditional transmission lines (wire pair, coaxial cable, waveguide) are purchased as
standard components and cut to length. Microstrip, and similarly fabricated line techno-
logies, however, are typically more integrated with the active and passive components that
they connect and require designing for each particular circuit application. A detailed descrip-
tion of microstrip is therefore given along with the design equations required to obtain
the physical dimensions that achieve the desired electrical characteristics, given constraints
such as substrate permittivity and thickness that are fixed once a (commercial) substrate
has been selected. The limitations of microstrip including dispersion and loss are discussed
and methods of evaluating them are presented. The problem of discontinuities is addressed
and models for the foreshortened open end (an approximate open circuit termination), vias
(an approximate short circuit termination), mitred bends (for reducing reflections at microstrip
corners) and T-junctions are described.
In addition to a simple transmission line technology for interconnecting active and passive
devices, microstrip can be used as a passive device technology in its own right. Microstrip
implementation of low-pass filters is described and illustrated with a specific example. The
general theory of coupled microstrip lines, useful for generalised filter and coupler design, is
presented and the concepts of odd- and even-modes explained. Equations and design curves
for obtaining the physical microstrip dimensions to realise a particular electrical design
objective are presented. The directivity of parallel microstrip couplers is discussed and
simplified expressions for its calculation are presented. Methods of improving coupler
Overview 5

performance by capacitor compensation are described. The discussion of practical microstrip


design methods concludes with a brief survey of other microstrip coupler configurations
including Lange couplers, branch-line couplers and hybrid rings.
Network methods represent a fundamental way of describing the effect of a device or sub-
system inserted between a source and load (which may be the Thévenin/Norton equivalent
circuits of a complicated existing system). From a systems engineering perspective, the net-
work parameters of the device or subsystem describe its properties completely – knowledge
of the detailed composition of the device/subsystem (i.e. the circuit configuration or values
of its component resistors, capacitors, inductors, diodes, transistors, transformers, etc.) being
unnecessary. The network parameters may be expressed in a number of different ways, e.g.
impedance (z), admittance (y), hybrid (h), transmission line (ABCD) and scattering (s) para-
meters, but all forms give identical (and complete) information and all forms can be readily
transformed into any of the others. Despite being equivalent, there are certain practical
advantages and disadvantages associated with each particular parameter set and at RF and
microwave frequencies these weigh heavily in favour of using s-parameters. A brief review
of all commonly used parameter sets is therefore followed by a more detailed definition and
interpretation of s-parameters for both one- and two-port (two- and four-terminal) networks.
The reflection and transmission coefficients at the impedance discontinuities of a device’s
input and output are described explicitly by the device’s s-parameters. One of the central
problems in RF and microwave design is impedance matching the input and output of a
device or subsystem with respect to its source and load impedances. (This problem may be
addressed in the context of a variety of objectives such as minimum reflection, maximum
gain or minimum noise figure.) The chapter therefore continues with an account of the
most widely used aids to impedance matching, namely the Smith chart and its derivatives
(admittance and immitance charts). These aids not only accelerate routine (manual) design
calculations but also present a geometrical interpretation of relative impedance that can
lead to analytical insights and creative design approaches. Both lumped and distributed
element techniques are described including the classic transmission line cases of single
and double stub matching. The treatment of matching ends with a discussion of broadband
matching, its relationship to quality factor (Q-) circles that can be plotted on the Smith chart,
and microstrip line transformers.
Chapter 3 closes with a description of network analysers – arguably the most important
single instrument at the disposal of the microwave design engineer. The operating principles
of this instrument, which can measure the frequency dependent s-parameters of a device,
circuit or subsystem, are described and the sources of measurement error are examined. The
critical requirement for good calibration of the instrument is explained and the normal
calibration procedures, including the technologies used to make measurements on naked
(unpackaged) devices, are presented.

1.4 Amplifiers
Virtually all systems need amplifiers to increase the amplitude and power of a signal. Many
people are first introduced to amplifiers by means of low frequency transistor and operational
amplifier circuits. At microwave frequencies amplifier design often revolves around terms
such as available power, unilateral transducer gain, constant gain and constant noise figure
circles, and biasing the transistor through a circuit board track that simply changes its width
6 Microwave Devices, Circuits and Subsystems for Communications Engineering

in order to provide a high isolation connection. This chapter aims to explain these terms and
why they are used in the design of microwave amplifiers.
Chapter 4 starts by carefully considering how we define all the power and gain quantities.
Microwave frequency amplifiers are often designed using the s-parameters supplied by the
device manufacturer, so following the basic definitions of gain, the chapter derives expressions
for gain working in terms of s-parameters. These expressions give rise to graphical representa-
tions in terms of circles, and the idea of gain circles and their use is discussed.
If we are to realise an amplifier, we want to avoid it becoming an oscillator. Likewise, if
we are to make an oscillator, we do not want the circuit to be an amplifier. The stability of
a circuit needs to be assessed and proper stability needs to be a design criterion. We look
at some basic ideas of stability, and again the resulting conditions have a graphical inter-
pretation as stability circles.
Amplifiers have many different requirements. They might need to be low noise amplifiers
in a sensitive receiver, or high power amplifiers in a transmitter. Some applications require
narrowband operation while some require broadband operation. This leads to different
implementations of microwave amplifier circuits, and some of these are discussed in this
chapter.
At microwave frequencies the capacitance, inductance and resistance of the packages
holding the devices can have very significant effects, and these features need to be considered
when implementing amplifiers in practice. Also alternative circuit layout techniques can be
used in place of discrete inductors or capacitors, and some of these are discussed. In dealing
with this we see that even relatively simple amplifier circuits are described by a large
number of variables. The current method of handling this amount of data and achieving
optimum designs is to use a CAD package, and an outline of the use of these is also given.

1.5 Mixers
Mixers are often a key component in a communication or radar system. We generally
have our basic message to send, for example, a voice or video signal. This has a particular
frequency content that typically extends from very low frequency, maybe zero, up to an
upper limit, and we often refer to this as the baseband signal. Many radio stations, TV
stations, and mobile phones can be used simultaneously, and they do this by broadcasting
their signal on an individually allocated broadcast frequency. It is the mixer circuit that
provides the frequency translation from baseband up to the broadcast frequency in the trans-
mitter, and from the broadcast frequency back down to the original baseband in the receiver,
to form a superheterodyne system.
A mixer is a non-linear circuit, and must be implemented using a nonlinear component.
Chapter 5 first outlines the operation of the commonly used nonlinear components, the
diode and the transistor. After that the analysis of these circuits are developed, and the terms
used to characterise a mixer are also described. This is done for the so-called linear analysis
for small signals, and also for the large signal harmonic balance analysis.
The currently popular transistor mixers are then described, particularly the signal and
dual FET implementations that are common in Monolithic Microwave Integrated Circuits
(MMICs). The designs of many mixers are discussed and typical performance characteristics
are presented. The nonlinear nature of the circuit tends to produce unwanted frequencies at
the output of the mixer. These unwanted terms can be ‘balanced’ out, and the chapter also
discusses the operation of single and double balanced mixer configurations.
Overview 7

1.6 Filters
Chapter 6 provides the background and tools for designing filter circuits at microwave
frequencies. The chapter begins with a review of two-port circuits and definitions of gain,
attenuation and return loss, which are required in the later sections. The various filter
characteristics are then described including low-pass, high-pass, band-pass and band-stop
responses along with the order number of the filter and how this affects the roll-off of the
gain response from the band edges.
The various types of filter response are then described: Butterworth (maximally flat within
the passband), Chebyshev (equal ripple response in the pass band), Bessel (maximally flat in
phase response) and Elliptic (equal ripple in pass band and stop band amplitude response).
The chapter then addresses the topic of filter realisation and introduces the concept of the
low pass prototype filter circuit, which provides the basis for the filter design concepts in the
remainder of the chapter. This has a normalised characteristic such that the 3 dB bandwidth
of the filter is at a frequency of 1 radian/s and the load impedance is 1 ohm. The four types
of filter response mentioned above are then considered in detail, with the mathematical
description of the responses given for each.
The chapter then continues with the detail of low pass filter design for any particular
value of the order number, N. The equivalent circuit descriptions of the filters are given for
both odd and even values of N and also for T- and Π-ladders of capacitors and inductors.
Analyses are provided for each of the four filter response types and tables of component
values for the normalised response of each is provided.
As these tables are only applicable to the normalised case (bandwidth = 1 radian/s and
the filter having a load impedance of 1 ohm), the next stage in the description of the filter
realisation is to provide techniques for scaling the component values to apply for any
particular load impedance and also for any particular value of low-pass bandwidth. The
mathematical relationships between these scalings and the effects on the component values
of the filter ladder are derived.
So far, the chapter has only considered the low pass type of filter, so the remainder of the
chapter considers the various transformations required to convert the low pass response into
equivalent high pass, band pass and band stop responses for each of the filter characteristic
types. This therefore provides the reader with all the tools and techniques to design a
microwave filter of low pass, high pass, band pass or band stop response with any of the four
characteristic responses and of various order number.

1.7 Oscillators and Frequency Synthesisers


Chapter 7 describes the fundamentals of microwave oscillator design, including simple active
component realisations using diodes and transistors. The chapter commences with an intro-
duction to solid-state oscillator circuits considered as a device with a load. The fundamental
approach is to consider the oscillation condition to be defined so that the sum of the device and
load impedances sum to zero. Since the real part of the load impedance must be positive, this
implies that the real part of the active device’s effective impedance must be negative. This
negative resistance is achieved in practice by using a negative resistance diode or a transistor
which has a passive feedback network. The active device will have a nonlinear behaviour
and its impedance depends on the amplitude of the signal. The balancing condition for the
zero impedance condition therefore defines both the frequency and amplitude of oscillation.
8 Microwave Devices, Circuits and Subsystems for Communications Engineering

The chapter continues with a description of diode realisations of negative resistance


oscillators including those based on IMPATT and Gunn devices. This section includes
example designs using typical diode characteristics, optimum power conditions and oscilla-
tion stability considerations.
Transistor oscillators are then considered. The fundamental design approach is to consider
the circuit to be a transistor amplifier with positive feedback, allowing the growth of any
starting oscillating signal. This starting signal is most likely to be from ever-present noise
in electronic circuits. The feedback circuit is resonant at the desired oscillation frequency,
so only noise signals within the bandwidth of the resonant circuit will be amplified and fed
back, the other frequencies being filtered out. The possible forms of the resonant feed-
back circuit are discussed, these include lumped L-C circuits, transmission line equivalents,
cavity resonators and dielectric resonators.
The standard topology of transistor feedback oscillators such as the Colpitts, Clapp and
Hartley configurations are described and analysed from a mathematical point of view. This
section concludes with a discussion of some of the Computer Aided Design (CAD) tools
available for the design and analysis of solid-state microwave oscillator circuits.
The next section of the chapter deals with the inclusion of voltage-controlled tuning of
the oscillator so that the frequency of oscillation can be varied by the use of a controlling
DC voltage. The main implementations for voltage-controlled oscillators (VCOs) use varactor
diodes and Yttrium Iron Garnets (YIGs). Varactors are diodes whose junction capacitance
can be varied over a significant range of values by the applied bias voltage. When used as
part of the frequency selective feedback circuit, variation of this diode capacitance will lead
to a variation in the oscillation frequency. YIGs are high-Q resonators in which the
ferromagnetic resonance depends (among other factors) on the magnetic field across the
device. This in turn can be controlled by an applied voltage. As well as having a high Q
value (and therefore a highly stable frequency), YIGs are also capable of a very wide range
of voltage control, leading to very broadband voltage control. Examples of practical VCO
design complete this section.
The next section considers the characterisation and testing of oscillators. The various
parameters used to specify the performance of a particular oscillator include: the oscillator
frequency, characterised using a frequency counter or spectrum analyser; the output power,
characterised using a power meter; stability and noise (in both amplitude and phase), which
can again be analysed using a spectrum analyser or more sophisticated phase noise measure-
ment equipment.
The final section of the chapter deals with phase-locked oscillators, which use phase-
locked loops (PLLs) to stabilise the frequency of microwave oscillators. A fundamental
description of PLLs is given, along with a consideration of their stability performance.
These circuits are then incorporated into the microwave oscillators using a frequency
multiplier and harmonic mixers so that the microwave frequency is locked on to a lower,
crystal stabilised, frequency so that the characteristics of the highly stable low frequency
source are translated on to the microwave frequency.
RF Devices: Characteristics and Modelling 9

2
RF Devices: Characteristics
and Modelling
A. Suarez and T. Fernandez

2.1 Introduction
Semiconductor transistors and diodes both exhibit a non-linear current and/or voltage input–
output characteristic. Such non-linearity can make the behaviour of these devices difficult
to model and simulate. It does enable, however, the implementation of useful functions such
as frequency multiplication, frequency translation, switching, variable attenuators and power
limiting. Transistors and some types of diodes may also be active, i.e. capable of delivering
energy to the system, allowing them to be used in amplifier and oscillator designs. Passive
non-linear responses are used for applications such as frequency mixing, switching or power
limiting.
The aims of this chapter are: (1) to give a good understanding of the operating prin-
ciples of the devices presented and to convey factual knowledge of their characteristics and
limitations so as to ensure their appropriate use in circuit design; (2) to present accurate
equivalent circuit models and introduce some efficient modelling techniques necessary for
the analysis and simulation of the circuits in which the devices are employed; and (3) to
present the most common applications of each device, illustrating the way in which their
particular characteristics are exploited.
The chapter starts with a revision of semiconductor physics, including the general prop-
erties of semiconductor materials and band theory that is usually used to explain the origin
of these properties. This is followed by a detailed description of the two most important
semiconductor devices, diodes and transistors, in their various RF/microwave incarnations.
In keeping with the practical circuit and subsystem design ethos of the courses on which this
text is based, significant emphasis is placed on the devices’ equivalent circuit models that
are necessary for both traditional and modern computer-aided design.

Microwave Devices, Circuits and Subsystems for Communications Engineering Edited by I. A. Glover, S. R. Pennock
and P. R. Shepherd
© 2005 John Wiley & Sons, Ltd.
10 Microwave Devices, Circuits and Subsystems for Communications Engineering

2.2 Semiconductor Properties


Solids may be divided into three principal categories: metals, insulators and semiconductors
[1, 2]. Metals consist of positive ions, surrounded by a cloud of electrons. Free electrons,
which are shared by all the atoms, are able to move under the influence of an electric field
at 0 K. Metals have only one type of charge carrier, the electron, conduction being due to
electron movement only. The concentration of electrons in an electrically neutral metal is
always approximately the same whatever the material or temperature, with values between
1022 and 1023 cm−3.
Insulators are crystalline structures in which electrons are bound closely together in covalent
bonds. Electrons in insulators do not move under the influence of an electric field at room
temperature, T0 (= 300 K).
Semiconductors are crystalline structures composed of valence-IV atoms, linked by covalent
bonds. They behave as insulators at 0 K and as good conductors at room temperature. The
absence of one electron leaves a hole in the covalent bond. When applying an electric field,
the filling of this vacancy by another electron, leaving, in turn, another hole, gives rise to
an apparent hole movement. A positive charge value may be associated with every hole.
Semiconductors have thus two types of charge carriers: electrons and the holes.

2.2.1 Intrinsic Semiconductors


Intrinsic semiconductors may have any of several different forms: single elements with valence
IV, such as silicon (Si), germanium (Ge), and carbon (C), and compounds with average valence
IV. Among the latter, binary III–V and II–VI associations are the most usual. For example,
Gallium Arsenide (GaAs), a III–V compound, is commonly used for microwave devices due
to its good conduction properties. In the semiconductor crystal every atom is surrounded by
four other atoms and linked to them by four covalent bonds (Figure 2.1).
In compound semiconductors these bonds are formed between the positive and negative
ions with four peripheral electrons and are called hetero-polar valence bonds. For instance,
in the case of GaAs, four covalent bonds are formed between the negative ions Ga− and the
positive ions As+.

+4 +4 +4

+4 +4 +4

+4 +4 +4

Figure 2.1 Covalent bonding in semiconductor crystal


RF Devices: Characteristics and Modelling 11

The charge movement in semiconductors may be due to thermal agitation or to the


application of an electric field. In the former case the movement is random, while in the
latter case carriers move systematically in a direction, and sense, that depends on the applied
electric field, E. Electrons are accelerated in the opposite sense to that of the electric field
until they reach a final drift velocity, vn, given by:

vn = − µ n E (2.1)

where µ n is a proportionality constant, known as the electron mobility. Mobility is defined as


a positive quantity so Equation (2.1) is in agreement with the electron movement in opposite
sense to the electric field. The holes are accelerated in the same sense as the applied field
until they reach a final drift velocity, vp, given by:

vp = µ p E (2.2)

where µp is hole mobility. Mobility therefore depends on the type of particle, electron or hole,
and on the material. For electrons, its value is 1500 cm2 V −1s−1 in Si and 8500 cm2 V −1s−1 in
GaAs. The higher mobility value makes the latter well suited for microwave applications.
For relatively low electric field strengths, the mobility µ has a constant value, resulting
in a linear relationship between the velocity and the applied electric field. At higher field
strengths, however, mobility depends on electric field, i.e. µ = µ(E), which gives rise to a
nonlinear relationship. For still higher values of electric field, saturation of the drift velocity
is observed.
Electron and hole movement due to the application of an electric field, i.e. drift current,
may be easily related to the applied field. The current density, Dn, due to the electron
movement, is given by:

dIn dQn 1
Jn = = ⋅ (2.3)
dS dt dS

where In is total electron current through the differential surface element dS and Qn is the
total electron charge passing the surface in differential time element dt. For a total electron
number N, the total charge will be Qn = −Ne, where e is the (magnitude of the) electron
charge, i.e. 1.602 × 10−19 C. The differential charge element dQn is given by dQn = −dNe.
Since the electron density per unit volume, n, is given by:

dN
n= (2.4)
dV

the element dN may be written as dN = ndV and substituting into Equation (2.3) gives:

dV 1
Jn = −ne = −nevn (2.5)
dt dS

Combining Equations (2.1) and (2.5):

Jn = µ nenE (2.6)
12 Microwave Devices, Circuits and Subsystems for Communications Engineering

By following the same steps a similar result can be obtained for the hole current, i.e.:

Jp = µ pepE (2.7)

where p is the hole concentration (i.e. number per unit volume). The total drift current is,
therefore:

Jd = (µ n n + µ p p)eE (2.8)

Ohm’s law can be expressed as:

J = σE (2.9)

where σ is conductivity (in S/m). By comparing Equations (2.8) and (2.9) it is possible to
derive an expression relating a material’s conductivity with its mobility values, i.e.:

σ = µ n ne + µ p pe (2.10)

It is clear from Equation (2.10) that increasing the carrier concentration enhances the
conductivity properties of the semiconductor. As the temperature rises, more covalent bonds
break and for every broken bond a pair of carriers, an electron and a hole, are generated.
The conductivity of the intrinsic material, therefore, increases with the temperature.
A second type of current, quite different in origin from the drift current discussed
above, may also exist in semiconductor devices. This diffusion current is due to any non-
homogeneity in the carrier concentration within the semiconductor material. In order to
balance concentrations, carriers move from the higher concentration regions to regions of
lower concentration, the movement being governed by the concentration gradient. For a
one-dimension device (i.e. a device in which variations only occur in one, the x, dimension),
electron diffusion current density, JDn, is given by [1, 2]:

dn
JDn = eDn (2.11)
dx

where Dn is the diffusion coefficient for electrons. Dn is related to electron mobility and
temperature through the expression [1, 2]:

kT
Dn = µn (2.12)
e

where k is Boltzmann’s constant (1.381 × 10−23 J/K) and T is absolute temperature.


For holes, the diffusion current density, JDp, is given by:

dp
J Dp = −eDp (2.13)
dx

where Dp is the diffusion coefficient for electrons given by:

kT
Dp = µp (2.14)
e
RF Devices: Characteristics and Modelling 13

In addition to the drift and diffusion currents, a third current component should be con-
sidered in the semiconductor when a time variable electric field is applied. This is displacement
current. Displacement current density, Jd, given by [1, 2]:

dE
J d = ε sc (2.15)
dt

where εsc is the semiconductor dielectric constant.

Self-assessment Problems
2.1 What are the mechanisms that respectively give rise to drift current and diffusion
current in a semiconductor material?

2.2 What is the relationship between semiconductor conductivity and carrier mobility?

2.3 Why is GaAs preferred to silicon for microwave circuit design?

2.2.2 Doped Semiconductors


As shown by Equation (2.10), semiconductor conductivity improves with carrier concentration.
This may be increased by adding impurities to (i.e. doping) the semiconductor. There are
two different types of doping: N-type, when valence-V atoms are added, and P-type, when
valence-III atoms are added.

2.2.2.1 N-type doping


A semiconductor material is N-doped, or N-type, when a certain concentration, ND, of
valence-V atoms, such as phosphorus (P), arsenic (As) or antimony (Sb), is introduced into
the crystal (Figure 2.2).

+4 +4 +4

+
+4 +5 +4

+4 +4 +4

Figure 2.2 N-doped semiconductor crystal


14 Microwave Devices, Circuits and Subsystems for Communications Engineering

+4 +4 +4

e
+4 +3 +4

+4 +4 +4

Figure 2.3 P-doped semiconductor crystal

These valence-V impurity atoms are known as donors. Due to the resulting lack of
symmetry, one of the five valence electrons in each impurity atom is linked only weakly to
the crystal lattice. These electrons therefore need only a small amount of energy to become
conductive. At room temperature, in fact, there is sufficient thermal energy to ionise virtually
all the impurity atoms. It is therefore possible to assume n ≅ ND at room temperatures, since
the impurity electrons are far more numerous than those created by thermal electron-hole
pair generation in the intrinsic material. In an N-type material, electrons are thus majority
carriers, while holes are minority carriers. As temperature increases, carrier creation by
thermal generation of (intrinsic material) electron-hole pairs becomes progressively more
important and at very high temperatures (500 to 550 K for Si) the material loses its doped
characteristics, behaving essentially as an intrinsic semiconductor.

2.2.2.2 P-type doping


A semiconductor material is P-doped when a certain concentration, NA, of valence-III atoms,
such as boron (B) or gallium (Ga), is introduced into the crystal (Figure 2.3).
These valence-III impurity atoms are known as acceptors. Due to the resulting lack of
symmetry, the impurity atoms easily accept electrons from the surrounding covalent bonds,
thus becoming ionised. For each ionised impurity, a hole is left in the crystal lattice. The
filling of this hole by an electron leaves in turn another hole, equivalent to a positive charge
displacement. The hole therefore behaves as a positive charge carrier. Measurements of
the ease with which holes can be made to move leads to their being ascribed an equivalent
mass value. At room temperature all the impurity atoms, with concentration NP, are usually
ionised and it is possible to make the approximation p ≅ NP. In this case, the holes are
the majority carriers, while the electrons, generated by thermal effects within the intrinsic
material, are the minority carriers. At very high temperature, due to the strong electron-hole
creation process, the material basically behaves as an intrinsic semiconductor.

2.2.3 Band Model for Semiconductors


It is well known that the electrons in an atom can neither have arbitrary energy values (but
instead can occupy only certain, discrete, allowed energy levels) nor can they share a single
RF Devices: Characteristics and Modelling 15

CONDUCTION BAND
W

e e e
WC

WF

WV

VALENCE BAND

Figure 2.4 Semiconductor energy band diagram

energy value. If two atoms are placed in close proximity, the number of allowed energy
levels is doubled in order to host twice the number of electrons without more than one
electron occupying a single level. If a large number of electrons are placed close together,
as is the case within a solid, the discrete levels become packed so tightly together as to
effectively give rise to a continuous allowed energy band. Forbidden energy bands then
separate permitted energy bands.
The atoms of an intrinsic semiconductor are, at 0 K, joined by covalent bonds established
between their four valence electrons and an equal number of electrons from each of the four
neighbouring atoms in the crystal. These electrons may be thought to occupy the discrete
levels of an energy band called the valence band. As the temperature rises, some of these
bonds may break. The corresponding electrons will be more energetic, being located at
higher energy levels. These higher levels, containing electrons capable of moving under the
effect of an electric field, form what is known as the conduction band (Figure 2.4).
An electron making the transition to the conduction band leaves a hole in the valence
band. Between the valence and conduction bands there is a forbidden energy gap. This
energy gap represents the energy needed to break a covalent bond. In good insulators, the
valence band is complete at room temperature and the conduction band is empty, with a
wide energy gap between the two. In metals, there is no forbidden band, the valence and
conduction bands overlap with the electrons, which are the only carriers, being located at the
lower levels of the non-saturated (i.e. only partly filled) conduction band.
The Fermi energy level is defined as the energy level at which the probability of finding
an electron is 0.5. For an intrinsic material, the Fermi level is always located in the middle
of the forbidden energy gap, whatever the temperature [1, 2]. (For 0 K, the probability of
finding an electron below the Fermi level is 1.0 and the probability of finding an electron
above the Fermi level is zero.) The Fermi level is always defined in the absence of an
external force, such as a bias generator. When an external force is present, the so-called
pseudo-Fermi level must be considered.
In the case of an N-doped material, one of the electrons in each impurity atom is far more
energetic than the other four, which form close valence bonds with the four surrounding atoms.
The more energetic electrons will, by definition, be located at higher energy levels in the
band diagram. These higher levels form what is called the donor band, located immediately
beneath the conduction band (Figure 2.5).
16 Microwave Devices, Circuits and Subsystems for Communications Engineering

W
CONDUCTION BAND

e e e e
WC
WD

WV

VALENCE BAND

Figure 2.5 Energy band diagram of N-doped semiconductor showing donor


band below conduction band

The donor band will be complete at 0 K. However, since these electrons only need a small
energy supply to become conductive, the donor band will be generally empty at room
temperature, all its electrons having moved to the conduction band. It must be noted that
these carriers do not leave holes, in contrast to the case for thermal generation that results
in electron-hole pairs. In N-doped semiconductors, the Fermi level is located immediately
below the conduction band [1, 2] and as temperature increases, it progressively descends
towards the middle of the forbidden band. This is consistent with the increasing number of
electron-hole pairs as the temperature rises, progressively making the material behave as an
intrinsic one. (For an intrinsic material, the Fermi level is always located in the middle of
the forbidden energy gap, whatever the temperature.)
In the case of a P-doped material, an acceptor band is located immediately above the
valence band (Figure 2.6).

W
CONDUCTION BAND

e e
WC

WA
WV

VALENCE BAND

Figure 2.6 Energy band diagram of P-doped semiconductor showing acceptor


band above valence band
RF Devices: Characteristics and Modelling 17

The acceptor band is empty at 0 K. As the temperature rises, however, electrons in the
valence bonds will move to the acceptor band, leaving holes in the valence band, ready
for conduction. For a P-doped material, at low temperature, the Fermi level is located
immediately above the valence band [1, 2], moving to the middle of the forbidden band as
the temperature rises.

Self-assessment Problems
2.4 Why does doping increase the conductivity of semiconductor material?

2.5 Why is the valence band of a semiconductor complete at 0 K?

2.6 Why is the donor band located immediately below the conductor band?

2.2.4 Carrier Continuity Equation


Charge within the semiconductor must be conserved and it follows that its variation with
time must equal the current flow, plus the charge creation and recombination loss per unit
time. Charge creation may be due to thermal effects, to the illumination of the material, or
to the application of a very high electric field. Charge recombination occurs when a free
electron, moving through the crystal lattice, fills a hole, both of them disappearing as free
charges. The average time between carrier creation and recombination is called the mean
carrier lifetime, τ. Perhaps surprisingly, the mean carrier lifetime is different for holes and
electrons.
In the case of holes, for a semiconductor surface S and differential length dx, the charge
loss per unit time (i.e. the current lost, dIrp) due to recombination will be [1, 2, 3]:

p
dIrp = − eSdx (2.16)
τp

where τp is mean hole lifetime. If, at the same time, gp holes are generated per unit volume
per unit time, the corresponding charge increase per unit time, i.e. the current gained,
will be:

dIgp = gpeSdx (2.17)

The differential element for the total hole current, due to drift and diffusion effects is then
given by:

∂J p
dIp = Sdx
∂x
 ∂2 p ∂( pE) 
=  −eDp 2 + eµ p  Sdx (2.18)
 ∂x ∂x 
18 Microwave Devices, Circuits and Subsystems for Communications Engineering

and the total charge variation with time, due to holes, is:

∂p
eSdx = dIrp + dIgp − dIp
∂t
p  ∂2 p ∂( pE) 
= − eSdx + g peSdx +  eDp 2 − eµ p  Sdx (2.19)
τp  ∂x ∂x 

In Equation (2.19) it is possible to eliminate the term eSdx, obtaining the following expression:

∂p p ∂2 p ∂( pE)
=− + g p + Dp 2 − µ p (2.20)
∂t τp ∂x ∂x

which is known as the hole continuity equation. A similar equation may be deduced for the
electrons, i.e.:

∂n n ∂2 n ∂(nE)
= − + gn + Dn 2 + µn (2.21)
∂t τn ∂x ∂x

Equations (2.20) and (2.21) govern much of semiconductor device physics.

2.3 P-N junction


A P-N junction is obtained when two pieces of semiconductor, one P-type and one N-type,
with similar impurity concentrations, are connected. Due to the difference in electron and
hole concentration on either side of the junction, electrons from atoms close to the junc-
tion in the N region will diffuse to the P side in an attempt to balance the concentrations.
Similarly, holes from atoms close to the junction in the P region will diffuse to the N side.
The currents due to both these processes have the same sense since they are due to charges
of opposite sign moving in opposite directions.

2.3.1 Thermal Equilibrium


The electrons moving from N to P material will leave fixed positive ions, while the holes
moving from P to N material will leave fixed negative ions (Figure 2.7).

P N
+ + − −
− − − − − + + + + +
+ + − −
− − − − − + + + + +
+ + − −
− − − − − + + + + +
+ + − −
− − − − − + + + + +
−Xp 0 Xn

WD

Figure 2.7 Schematic representation of P-N junction


RF Devices: Characteristics and Modelling 19

A depletion zone, void of mobile charge is thus created close to, and on both sides of, the
junction. This zone is also known as the space-charge region. The fixed charges give rise
to an electrostatic potential, opposing the diffusion process, which drastically reduces the
current magnitude. At the same time the few minority carriers (electrons in the P region and
holes in the N region) created thermally, will be attracted by the fixed charges. Both P and
N minority currents add, the total minority current having opposite sense to the current due
to majority carriers. Equilibrium is attained when the total current reaches zero. The final
value of the potential barrier, also called built-in potential, is denoted by φ 0. Its magnitude
depends on the semiconductor material and on the doping level, with typical values between
0.5 and 0.9 V.
For a constant acceptor concentration Na in the P region and a donor concentration Nd in
the N region, the fixed charge density along device is shown in Figure 2.8(a).
In Figure 2.8 the junction plane has been taken as the x co-ordinate origin. The limits of
the depletion region are defined as −xp on the P side and xn on the N side. From the principle

P Depletion N

ND
+ + +
E + + +

(a)
−Xp − − Xn
− −
− −
−NA

E
dE ρ
=
dX εscε 0
(b)

−Em
V

−dV
φ0 E=
dX
(c)

Figure 2.8 Charge (a), field (b) and potential (c) variations across a P-N junction
20 Microwave Devices, Circuits and Subsystems for Communications Engineering

of electrical neutrality, the absolute value of the total charge at both sides of the junction
must be equal, i.e.:

Na exp S = Nd exn S (2.22)

Therefore:

Na xp = Nd xn (2.23)

According to Equation (2.23), the extension of the space-charge zone is inversely pro-
portional to the doping concentration. Its extension will therefore be smaller at the side of
the junction with the highest doping concentration. In the derivation of Equation (2.23) an
abrupt junction has been considered, with a constant impurity concentration on both sides of
the junction. Other doping profiles are possible with a doping concentration (including the
carrier sign) N(x). A linear profile, for example, will have a constant slope value and will
satisfy N(0) = 0.
The electric field E along the depletion zone can be determined from Poisson’s equation,
i.e.:

dE ρ
= (2.24)
dx ε sc
where ρ is charge density (in C/m3) and εsc is semiconductor permitivity. By integrating
Equation (2.24) a field variation curve (Figure 2.8(b)) is obtained.
Finally, the potential, V, may be calculated using E = −dV/dx. Integrating both sides leads
to the curve in Figure 2.8(c). The potential variation along the space-charge region leads to
a bending of the energy bands, due to the relationship W = −eV, where W is the energy. The
resulting energy diagram is shown in Figure 2.9.

CONDUCTION BAND
E Cp

eφ0

E Cn

EF EF
EVp
VALENCE BAND

E Vn
DEPLETION

Figure 2.9 Energy band diagram for a P-N junction


RF Devices: Characteristics and Modelling 21

Since no external bias is applied, the Fermi level must be constant along the device
(which represents another explanation for the band bending). According to Figure 2.9,
electrons from the N region and holes from the P region now have great difficulty in
traversing the junction plane, and only a few will surmount the resulting energy barrier. The
minority electrons in the P region and the minority holes in the N region, however, traverse
the junction without opposition. As previously stated, the value of the total current, once
equilibrium has been reached, is zero.

2.3.2 Reverse Bias


A P-N junction is reverse-biased when an external voltage V is applied with the polarity as
shown in Figure 2.10.
This increases the potential barrier between the P and N regions to the value φ0 + V
(Figure 2.11), which reduces majority carrier diffusion to almost zero.
The increase in height of the potential barrier does not affect the minority carrier current,
since these carriers continue to traverse the junction without opposition. A small reverse
current Is is then obtained, which constitutes the major contribution (under reverse bias

P N

Figure 2.10 P-N junction in reverse bias

CONDUCTION BAND
E Cp

e(φ0 + V)

EFp E Cn
eV
EVp EFn
VALENCE BAND

E Vn
DEPLETION

Figure 2.11 Increased potential barrier of P-N junction in reverse bias


22 Microwave Devices, Circuits and Subsystems for Communications Engineering

conditions) to the total device current. Is is called the reverse saturation current, its size
depending on the temperature and the semiconductor material.
Due to the device biasing, no absolute, Fermi level can be considered, and pseudo-Fermi
levels are introduced instead, one on each side of the junction. In terms of potential, the differ-
ence between these pseudo-Fermi levels, VFp − VFn, equals the applied voltage V, with V > 0.
As a result of the reverse bias, the width of the depletion region increases compared to
that obtained for thermal equilibrium. This is due to electrons moving towards the positive
terminal of the biasing source and holes moving towards the negative terminal. The width
of the depletion region, given by wd = xn + xp, therefore depends on the applied voltage,
i.e. wd = wd (V).
The depletion region, composed of static ionized atoms, is non-conducting and can be con-
sidered to be a dielectric material. This region is bounded by the semiconductor neutral zones,
which, due to the doping, are very conducting. This structure therefore exhibits capacitance,
the value of which is known as the junction capacitance, Cj. Junction capacitance can be
calculated from:

S
C j (V ) = ε sc (2.25)
wd (V )

where S is the effective junction area and the dependence on applied voltage, V, is explicitly
shown. From Poisson’s equation, it is possible to obtain the following expression for the
width of the depletion zone, wd, as a function of the applied voltage V [1, 2]:

Cj 0
C j (V ) = γ (2.26)
 V
1 − 
 φ0 

where Cj0 is the capacitance for V = 0, γ = 0.5 for an abrupt junction and γ = 0.33 for a
graded (i.e. gradual) junction.
As reverse bias is increased, a voltage V = Vb is eventually attained, for which the electric
field in the depletion zone reaches a critical value, Ec. At this (high) value of electric field,
electrons traversing the depletion region are accelerated to a sufficiently high velocity to
knock other electrons out of their atomic orbits during collisions. The newly generated
carriers will in turn be accelerated and will release other electrons during similar collisions.
This process is known as avalanche breakdown. The breakdown voltage, Vb, imposes an
upper limit to the reverse voltage that can normally be applied to a P-N junction.
The breakdown voltage, Vb, can clearly be found from [1, 2]:

Vb = Ecwd (2.27)

The breakdown voltage is thus directly proportional to the width of the depletion zone, wd.
The energy threshold required by an electron or a hole in order to create a carrier pair is
higher than the width of the forbidden energy band. The new carriers therefore have a non-
zero kinetic energy after their creation. The ionisation coefficient, α, gives the number of
pairs that are created by a single accelerated carrier per unit length. This coefficient depends
on the applied field and has a lower value for a larger width of the forbidden energy band.
RF Devices: Characteristics and Modelling 23

P N

Figure 2.12 P-N junction in forward bias

2.3.3 Forward Bias


For forward bias, the external generator is connected as shown in Figure 2.12.
This reduces the potential difference along the junction (Figure 2.13) to a net value of
φ0 − V.
This reduction of the potential barrier favours the majority carrier current that will grow
exponentially as V is increased. Actually, as V → φ0, the majority carrier current will be
limited only by the device resistance and the external circuit. The device resistance is due to
the limited conductivity value of the neutral zones (ρ = 1/σ). The minority carrier current, Is,
depends only on device temperature and is not affected by the biasing. The current variation
as a function of the applied voltage is modelled using [1, 2]:

I(V) = Is(eαV − 1) (2.28)

where the parameter α is given by [1, 2]:

e
α = (2.29)
kT

and where k is Boltzmann’s constant expressed as 8.6 × 10−5 eV/K.


Since an external bias is applied, pseudo-Fermi levels must be considered and their
potential difference will be now VFp − VFn = −V, with V > 0 (Figure 2.13).

e(φ0 − V)
CONDUCTION BAND

EFn
eV
EFp

VALENCE BAND

DEPLETION

Figure 2.13 Decreased potential barrier of P-N junction in forward bias


24 Microwave Devices, Circuits and Subsystems for Communications Engineering

Due to the limited carrier lifetime, electrons entering the P region only penetrate a certain
distance (the diffusion length, Ln) before recombining with majority holes. The same applies for
holes entering the N region, which only penetrate a distance Lp. Both diffusion lengths are
related to the respective diffusion constants and carrier lifetimes through the formulae [1, 2]:
Ln = Dnτ n
(2.30)
Lp = Dpτ p

Under forward bias conditions the depletion region is narrow, with a width, wd, much smaller
than the diffusion lengths, Ln and Lp. This is responsible for a charge accumulation on both
sides of the junction, represented by electrons on the P side, and holes on the N side. This
charge accumulation, which varies with applied voltage, gives rise to a non-linear capacitance
(recall the relationship C = dQ/dV). The value of this capacitance is therefore given by:

dQ dI
Cd = =τ = τα IS eαV (2.31)
dV dV
τ being the mean carrier lifetime. Cd is known as the diffusion capacitance. It is typically
observed in bipolar devices (having both electron and hole conduction), as a direct con-
sequence of conduction through minority carriers.

2.3.4 Diode Model


The electric model for the P-N diode consists of a current source (with the non-linear I–V
characteristic given by Equation (2.28)) in parallel with two capacitors; one for the junction
capacitance, Cj, and the other for the diffusion capacitance, Cd (Figure 2.14).
The model is completed with a series resistor, Rs, accounting for the loss in the neutral
regions. The diffusion capacitance often prevents P-N diodes from being used in microwave
applications due to the extremely low impedance value associated with it in this frequency
range.
Lp

Rs

+ Cp

Cj(V) I(V) Cd(V) V

Figure 2.14 Diode equivalent circuit model


RF Devices: Characteristics and Modelling 25

Ohmic contact
P+
Depletion
N

N+ Substrate

Ohmic contact

Figure 2.15 Schematic diagram of P-N junction structure

In order to connect a P-N diode in a conventional circuit, metal contacts at the device
terminals are necessary. These are fabricated, at each end of the device, by depositing a
metallic layer over a highly doped semiconductor section. The intense doping is usually
represented by a superscript +, i.e. P + on the P side of the junction and N+ on the N side. The
reasons why high doping is necessary are given later.
The final diode may be produced in chip or packaged forms. When the device is packaged,
two parasitic elements, due to the packaging, must be added to the model. One element is
a capacitance, Cp, which arises due to the package insulator, and the other element is an
inductance, Lp, which arises due to the bonding wires.

2.3.5 Manufacturing
P-N diodes are manufactured in mesa¬ [1, 2]. In this technology the active layers of the
device are realised by locally attacking the semiconductor substrate, but leaving some isolated
(i.e. non-attacked) regions. The attack may be chemical or mechanical. In P-N junctions, an
N layer is deposited by epitaxial growth over a silicon substrate N+ (Figure 2.15).
Acceptor impurities are then diffused over the surface, obtaining a P +N junction. Metal
contacts are then deposited at the two device ends, over the P + layer and below the N+ layer.
As will be seen later, the high doping level is necessary in order to avoid the formation of
Schottky junctions.

Self-assessment Problems
2.7 What is the reason for the formation of the potential barrier in the unbiased P-N
junction?

2.8 Why does the P-N junction mainly behave as a variable capacitance for reverse
bias voltage? What is the phenomenon that gives rise to this variable capacitance?

2.9 Why is the diffusion capacitance due to the bipolar quality of the P-N junction diode?
26 Microwave Devices, Circuits and Subsystems for Communications Engineering

2.3.6 Applications of P-N Diodes at Microwave Frequencies


At microwave frequencies the diffusion capacitance, Cd, severely limits P-N diode applica-
tions which depend on its non-linear current characteristic. Most microwave applications
depend instead on the non-linear behaviour of the junction capacitance, Cj. These are known
as varactor applications.
The junction capacitance has voltage dependence given by:

Cj 0
C j (V ) = γ (2.32)
 V
1 − 
 φ0 

with γ = 0.5 for an abrupt junction and γ = 0.33 for a gradual junction (i.e. a junction with
linear spatial variation of the impurity concentration).
The capacitance Cj0 depends on the semiconductor material, on the doping concentrations,
Na and Nd, and on the P-N diode junction area S. Its value is usually between 1 pF and
20 pF. Varactor diodes are often used as variable capacitors to modify the resonant frequency
of a tuned circuit. In order to achieve the largest variation range, a high ratio between the
maximum and minimum capacitance values must be guaranteed. According to Equation
(2.32), the highest capacitance value of the varactor diode (for reverse-bias) is Cj0. Minimum
values are obtained for high reverse bias, but care must be taken to avoid avalanche break-
down. The quality factor, Q, of a varactor diode, which is a measure of the capacitive
reactance compared to series resistance (or stored energy to energy loss per cycle), is given
by [4]:

1
Q(V ) = (2.33)
2π f RsC j (V )

where f is frequency. As is clear from Equation (2.33) quality factor has a sensitive frequency
dependence. A silicon device, for example, with a bias voltage V = −4 volts, has a Q factor
of 1800 at 50 MHz and 36 at 2.5 GHz. Q factors are much higher in GaAs devices due to
the higher mobility value, which implies a lower Rs. The frequency, fc, at which the Q takes
the value unity is often taken as a device figure of merit [4].
The following example illustrates how the electrical model of a varactor diode can be
obtained.

Example 2.1

A packaged varactor diode, with a negligible loss resistance, can be modelled by the circuit
shown in Figure 2.16.
The total diode capacitance is measured as a function of reverse bias at 1 MHz, resulting
in the curve shown in Figure 2.17.
When the diode is biased at V = −20 volt, the input impedance exhibits a series resonance
at fr = 10 GHz. The P-N junction is of an abrupt type, with φ0 = 0.6 V. Determine the value
of the model elements.
RF Devices: Characteristics and Modelling 27

Lp

Cj(V) Cp

Chip Package

Figure 2.16 Equivalent circuit model of a varactor diode

4.5

4
Total Capacitance (pF)

3.5

2.5

1.5

0.5
–40 –35 –30 –25 –20 –15 –10 –5 0
Diode Voltage (V)

Figure 2.17 Diode capacitance as a function of reverse bias at 1 MHz

The capacitance has been intentionally measured at low frequency, in order to ensure that
the inductance is negligible. Two values are taken from the curve:

CT (20) = Cj (20) + Cp = 0.981 pF

CT (30) = Cj (30) + Cp = 0.843 pF

Then:

Cj (20) − Cj (30) = 0.138 pF

Using Equation (2.32):

Cj0 = 4.5 pF
28 Microwave Devices, Circuits and Subsystems for Communications Engineering

For V = −20 volt, the intrinsic diode capacitance will be:

4.5
C j (20) = 1/ 2
= 0.767 pF
 20 
1+
 0.6 

Therefore:

Cp = 0.213 pF

Since the series resonance occurs at fr = 10 GHz, then:

1
Lp = = 0.258 nH
CT (20)(2π fr )2

Common applications of varactor diodes include amplitude modulators, phase-shifters


and frequency multipliers. Some of these are analysed below.

2.3.6.1 Amplitude modulators


Consider the schematic diagram in Figure 2.18 showing a transmission line with a varactor
diode and an inductor connected in parallel across it.
The capacitor Cb is a DC-block, used for isolating the diode biasing [4]. Let f0 be the
frequency of the microwave signal at the transmission line input. The inductor is selected in
order to resonate with the varactor capacitance at f0, for a certain bias voltage V1, i.e.:

1
2π f0 = (2.34)
LC j (Vi )

The impedance presented to the microwave signal by the parallel circuit will be infinite for
V = V1. For this bias voltage, the signal will therefore be transmitted unperturbed along the
transmission line. If the diode voltage is now changed to a small value, V2 (around zero volts),
the diode capacitance will increase, presenting a low value of impedance and thus reduce the
transmitted amplitude of the microwave signal.

Cb
Input L VDC < 0 Output

Figure 2.18 Equivalent circuit of varactor diode connected in parallel across a transmission line
RF Devices: Characteristics and Modelling 29

Input port 1 3 Output port

Figure 2.19 Implementation of phase shifter using a circulator and a varactor diode

The circuit shown in Figure 2.18 can be used as an amplitude modulator by biasing the
diode at the average value VDC = (V1 + V2)/2 and superimposing the modulating signal vm(t).

2.3.6.2 Phase shifters


Varactor diodes may be used as phase shifters by taking advantage of their variable reactance.
A possible topology, based on a circulator, is shown in Figure 2.19.
For an ideal circulator, all the input power from port 1 will be transferred to port 2. If the
diode resistance loss is neglected, complete reflection will occur at the purely reactive diode
termination, the value of reactance will affect the phase of the reflected signal, however. By
varying the varactor bias the phase shift can therefore be modified. Finally, the reflected
signal is transferred to port 3.

Example 2.2
Determine the phase shift provided by the circuit of Figure 2.19 for input frequency 10 GHz
and two bias voltages of the varactor diode: V1 = −0.5 V and V2 = −6 V. Obtain a general
expression for the phase shift θ versus the varactor bias. Assume an ideal circulator and
varactor parameters Cj0 = 0.25 pF, γ = 0.5 and φ0 = 0.8 V.
The varactor capacitance variation, as a function of the applied voltage, is given by:

Cj 0 0.25
C j (V ) = 1/ 2
= 1/ 2
pF
 V  V 
1 −  1−
 φ0   0.8 

Therefore:

C(−0.5) = 0.1961 pF

and:

C(−6) = 0.0857 pF
30 Microwave Devices, Circuits and Subsystems for Communications Engineering

Considering the varactor impedance as purely reactive, the reflection coefficient at the diode
terminals will be:

jXin − Z0
Γ=
jXin + Z0

where:

−j
jXin =

Replacing the two capacitance values:

Xin(−0.5) = −81.2 Ω

Xin(−6) = −185.7 Ω

Substituting into the expression for the reflection coefficient:

Γ(−0.5) = e−j1.10

Γ(−6) = e−j0.53

In order to obtain a general equation for the phase shift, the following expressions are
considered:

−Z0 + jXin = Ae jα

Z0 + jXin = Ae−j(α −π )

The phase of the reflection coefficient is then given by:

1
θ = 2α − π = 2 arctg
C j (V )ω 0 Z0

2.3.6.3 Frequency multipliers


The non-linear response of varactor capacitance may be used for harmonic generation.
Consider the circuit shown in Figure 2.20.
Assume that the voltage at the diode terminals is a small amplitude sinusoid superimposed
on the bias voltage VDC, i.e.:

v(t) = VDC + V1 cos(ω t) (2.35)

The non-linear capacitance will then vary as:

Cj 0
C j (t ) = γ (2.36)
 (VDC + V1 cos(ω t )) 
1 − φ0 
 
RF Devices: Characteristics and Modelling 31

ω 2ω

1
4
4
2
4
4
3

1
4
4
2
4
4
3
L1 C1 L2 C2

Z0 +
VDC + V1 cos ωt
Z0

Eg

Figure 2.20 Circuit for harmonic generation using varactor diode

Assuming small RF variations about the bias point, the preceding expression may be developed
in a first-order Taylor series about VDC giving:

Cj (t) = Cj (VDC)[1 + a cos(ω t)] (2.37)

where a is given by:

γ V1
a= (2.38)
φ0 − VDC

The current through the diode will be:

dv
I j (V ) = C j (V ) (2.39)
dt

Differentiating the sinusoidal voltage:

I(t) = Cj (VDC)[1 + a cos(ω t)]V1ω [−sin(ω t)] (2.40)

Using trigonometric identities:

1
I(t) = −Cj (VDC)V1ω sin(ω t) − [Cj (VDC)aV1ω] sin(2ω t) (2.41)
2

The ratio between the second harmonic component and the fundamental one is given by:

I2 a 1 γ V1
= = (2.42)
I1 2 2 φ0 − VDC

Finally, the series resonant circuit (with resonant frequency 2ω) selects the second harmonic
component, resulting in the desired frequency doubling action.
32 Microwave Devices, Circuits and Subsystems for Communications Engineering

2.4 The Schottky Diode


When two materials with different band structures are joined together, the resulting junction
is called a hetero-junction. Such junctions may be of two main types: semiconductor-
semiconductor or metal-semiconductor.
The Schottky junction is a metal-semiconductor hetero-junction. Its working principle is
similar to that of a P-N junction, but without the charge accumulation problems of the latter,
which enables the use of its non-linear current characteristic for forward bias applications
such a rectification, mixing and detection. Other differences with the PN junction are the
lower value of built-in potential (φ0 ≅ 0.4 V), the higher reverse current, Is, and the lower
avalanche breakdown voltage, Vb.
Ohmic contacts represent another form of metal-semiconductor hetero-junction. The current-
voltage characteristic resulting from a metal-semiconductor junction may, in fact, be Schottky
(rectifying) or ohmic, depending on the semiconductor doping level. The former is obtained
for a doping level lower than 1017 cm−3. The latter is obtained for a doping level higher
than 1018 cm−3. For this high doping level, the current-voltage characteristic degenerates to
(approximately) a straight line resulting in the ohmic contact [2].
The band structure of a metal is very different from that of a semiconductor. The free
electrons (in a very large numbers) occupy the lower levels of a conduction band that
is not saturated, i.e. that contains more allowed levels than those occupied. Considering
an N-type semiconductor, before metal and semiconductor come into contact, the free
energy level, E0, is the same for both. The respective Fermi levels, however, are different
(Figure 2.21).
The Fermi level is lower in the metal, since it has a lower occupation density of energy
levels. Three cases are analysed below for the Schottky junction. These cases are thermal
equilibrium, reverse bias and forward bias.

2.4.1 Thermal Equilibrium


Although the number of electrons is much higher in the metal than in the N-type semicon-
ductor, the occupation density of energy levels is higher for the semiconductor, since there are
less available levels. When the metal and the semiconductor come into contact, the electrons
from the semiconductor tend to diffuse into the metal but not the converse. Electrons from
the semiconductor atoms that are close to the junction plane will cross it, leaving positively
ionised donors (Figure 2.22(a)).

EO EO
φn
φm
EC
EFn
EFm
Ev
(a) (b)

Figure 2.21 Comparison of Fermi levels in (a) metal and (b) N-type semiconductor
RF Devices: Characteristics and Modelling 33

+ + N
Metal + + (a)
+ + Semiconductor

χ EO
φ0

EC (b)
EF

ρ
ND
(c)
X

(d)
X

Emax

Figure 2.22 Metal semiconductor junction (a) and the resulting energy band diagram (b),
charge density (c) and electric field (d)

This static charge is compensated for with electrons from the metal, which gives rise to a
static negative charge over the junction plane, not contributing to conduction. A space charge
zone therefore appears on both sides of the junction, giving rise to a built-in potential, φ0,
opposing electron movement from the semiconductor to the metal. This electron current
(noticeably reduced by the built-in potential) is compensated for by the small, thermally
generated, electron current flowing from the metal to the semiconductor. Since the free
electron density is so high in the metal, the space charge zone has a very short extension in
the metal. As expected, in thermal equilibrium, the global current is zero.
From the point of view of energy bands, once thermal equilibrium is attained, the
Fermi level must be of equal height across the entire structure. This leads to band bending
(Figure 2.22(b)). The same result is obtained when calculating the potential, V, along the
device. (As in the case of the P-N junction, V can be found by calculating the electric field
using Poisson’s equation and integrating.) Figure 2.22(c) and Figure 2.22(d) show the charge
density and electric field that exist in the junction region.
The energy diagram of Figure 2.22(b) makes clear the origin of the difficulty that semi-
conductor electrons experience in traversing the junction plane, as they must surmount the
potential barrier. The few electrons moving from the metal to the semiconductor simply
descend the potential barrier, attracted by the fixed positive ions. Due to the reduced space
charge zone on the metal side of the junction, band bending will be negligible in this region.
34 Microwave Devices, Circuits and Subsystems for Communications Engineering

2.4.2 Reverse Bias


For reverse biasing of a Schottky junction the external voltage source must be connected as
shown in Figure 2.23.
This gives rise to an increase in the height of the potential barrier, which now takes the
value φ0 + V (Figure 2.24).
This reduces to almost zero the number of semiconductor electrons moving into the metal.
The movement of electrons from the metal to the semiconductor is not affected, giving rise
to the reverse saturation current Is. This current has a bigger magnitude than in P-N junc-
tions. Due to the application of an external bias, pseudo-Fermi levels must be considered.
For reverse bias, their potential difference is given by VFm − VFn = V, with V > 0.
The space charge zone, containing the positively ionised donors, is equivalent to a dielectric
material, bounded by two very conductive regions: the metal and the neutral zone of the
N-type semiconductor. The width of the space charge zone varies with the applied reverse
voltage, giving rise to a variable junction capacitance. This capacitance has a value given by
Equation (2.32) with γ = 0.
When the reverse voltage is too high, the electrons from the metal traversing the depleted
zone will be accelerated enough to knock other electrons out of their atomic orbits. When this
process cascades, the current grows very rapidly with voltage. This effect is, of course, identical

N
Metal
Semiconductor

Figure 2.23 Reverse-biased Schottky junction

EO

EFm (φ0 + V)
V EC
EFn

EV

DEPLETION

Figure 2.24 Energy band diagram for a Schottky junction


RF Devices: Characteristics and Modelling 35

to that observed in reverse-biased P-N junctions as discussed in Section 2.3.2. A value of


breakdown voltage (Vb) and critical field strength (Ec), related via Equation (2.27), can there-
fore be defined. (In this context wd in Equation (2.27) is the width of the space charge zone.)
The breakdown voltage will be smaller as the doping increases, which is explained by
the existence of a larger number of free electrons available for the cascaded process. On the
other hand, as can be seen from Equation (2.27), Vb will be higher for a wider space charge
zone.
Schottky diodes have breakdown voltages that are lower than those observed in P-N
junctions and, as a consequence, smaller capacitance variation ranges may be expected in
Schottky varactors. When a Schottky diode is manufactured for varactor applications, the
doping is usually reduced, thus increasing Wd. This has the drawback, however, of increasing
the device loss resistor, Rs, since the semiconductor conductivity (related to the doping level
through the formula σ = µen) decreases. Doping reduction therefore decreases diode quality
factor (see Equation (2.33)). For non-linear current applications device resistance must be
reduced and diode doping is therefore increased (at the expense of decreased Vb).

2.4.3 Forward Bias


When connecting an external voltage source as shown in Figure 2.25, the Schottky diode
will be forward biased.
This reduces the height of the potential barrier to φ0 − V (Figure 2.26).

N
Metal
Semiconductor

Figure 2.25 Forward-biased Schottky junction

e(φ0 − V)
EO

EC
EFn
E Fm V

DEPLETION

Figure 2.26 Energy band diagram for forward-biased Schottky junction


36 Microwave Devices, Circuits and Subsystems for Communications Engineering

The difference between the pseudo-Fermi levels is now VFm − VFn = −V, with V > 0. As
V → φ0, the device current is limited only by the loss resistance and the external circuit.
Like the forward biased P-N junction, the current-voltage characteristic can be modelled
by [1, 2]:

I(V) = Is(eα V − 1) (2.43)

The parameter α, here, is given by [2]:

e
α = (2.44)
nkT

where n is an empirical factor, used to fit the model to the characteristic measured in


practice.
It is important to notice that, since the Schottky diode is a majority carrier device, no
charge accumulation takes place under forward bias conditions, there being no diffusion
capacitance. The Schottky diode is therefore better suited than the P-N junction to applica-
tions depending on the current-voltage characteristic.

2.4.4 Electric Model


The electric model for the Schottky diode comprises a non-linear current source I(V), in
parallel with a variable junction capacitance Cj , this parallel combination connected in series
with a loss resistor Rs (Figure 2.27), the latter, as usual, being due to the limited conductivity
of the semiconductor neutral zone. In comparison with the P-N diode model, the absence of
the diffusion capacitance should be noted.
For chip Schottky diodes, a series inductance Lp, accounting for the bonding wire, must be
considered. In the case of packaged devices, the parallel capacitance Cp, due to the package
insulator, must also be included.

Lp

Rs

+ Cp

Cj(V) V I(V)

Figure 2.27 Equivalent circuit model for Schottky junction


RF Devices: Characteristics and Modelling 37

Metal Oxide

N+ Substrate

Figure 2.28 Schottky junction structure

2.4.5 Manufacturing
Schottky diodes are manufactured in Si or GaAs. The bulk semiconductor is a highly doped
N+ substrate, with a width of several microns. Over this substrate, a thin N layer, of the
desired width, is obtained either through epitaxial growth or by ionic implantation. This
layer is protected by means of an oxide, as shown in Figure 2.28.
The metallic strip is realised by vacuum evaporation, through special holes in the oxide.
Suitable metals are aluminium and gold. The metallic deposit acts as one of the device
terminals. The second terminal is realised using an ohmic contact at the semiconductor base.

Self-assessment Problems
2.10 Before a metal and an N-type semiconductor are joined to form a Schottky
junction, why is the Fermi level of the metal located at a lower energy level than
that of the N-type semiconductor?

2.11 What is the phenomenon that gives rise to the negative static-charge accumulation
at the metal side of the Schottky junction?

2.12 Why do Schottky diodes not present a diffusion capacitance when forward
biased?

2.4.6 Applications
The application of Schottky diodes as varactors is similar to that of P-N junctions although,
due to the smaller breakdown voltages of the former, smaller capacitance variation ranges
are generally obtained.
Other applications of the Schottky diode stem from the non-linear characteristic of its
equivalent current source. Among them, applications as detectors and frequency mixers are
the most usual.
38 Microwave Devices, Circuits and Subsystems for Communications Engineering

2.4.6.1 Detectors
A detector can be used to demodulate a microwave signal or to measure its power and
microwave detectors are often based on the use of a Schottky diode. Such detectors take
advantage of the Schottky’s non-linear current source characteristic given by:
i(v) = Is(eα v − 1) (2.45)
where v (= v(t)) is the (signal) voltage across the diode terminals.
Considering a bias voltage V = V0 and a non-modulated microwave signal, the voltage v is
given by:

v = V0 + V cos ω t (2.46)
where V and ω are, respectively, the microwave signal amplitude and frequency. Provided V
is small, it is possible to develop i(v) as a Taylor series around the bias point, i.e.:

di 1 d 2i
i(v) = i(V0 ) + ∆v + ∆v 2 + . . . (2.47)
dv v0 2 dv 2 v0

where ∆v = V cos ω t. The derivatives are:

di
= α IseαV0 ≡ d1
dv v0
(2.48)
d 2i
= α 2 IseαV0 ≡ d2
dv 2 v0

Therefore, i(v) can be expressed as:


i(v) = I0 + d1∆v + (d2 /2)∆v 2 + . . . (2.49)

where I0 = i(V0). Replacing the ∆v value:

d2 2
i(v) = I0 + d1V cos(ω t) + V cos2(ω t) + . . . (2.50)
2

and using the trigonometric identity:

1
cos2ω t = (1 + cos 2ω t) (2.51)
2

gives:

d2 2 d
i(v) = I0 + V + d1V1 cos ω t + 2 V 2 cos 2ω t + . . . (2.52)
4 4
After low pass filtering, a (quasi) DC output may be obtained, which (apart from the offset
I0) is proportional to the square of the microwave signal amplitude V.
The preceding analysis is for the case of a quadratic detector. For larger microwave signal
amplitudes DC contributions from other terms in the Taylor series would result in a detector
RF Devices: Characteristics and Modelling 39

(DC) output that is proportional to the microwave signal amplitude [3, 4, 5], rather than its
power. For still larger signal amplitudes, a saturated behaviour is obtained. On the other hand,
for very small microwave signal amplitudes, the detected output would deviate negligibly
from I0 resulting in a noisy operating range if detection is possible at all [3, 4, 5].

2.4.6.2 Mixers
The mixing applications of the Schottky diode also arise from its non-linear current-voltage
characteristic. Using a Taylor expansion, this characteristic can be expressed in the form of
a power series, i.e.:

∆i = a1∆v1 + a2∆v2 + a3∆v3 + . . . (2.53)

where ∆i and ∆v are respectively the RF increments of current and voltage about the DC
bias point (V0 and I0).
In a mixer, the input voltage is composed of two tones: one corresponding to the input
signal, modelled here by Vin sin ωin, and the other corresponding to the local oscillator with
amplitude VLO and frequency ω LO radian/s, i.e.:

∆v = Vin sin(ω int) + VLO sin(ω LOt) (2.54)

Substituting this expression for ∆v into Equation (2.53) gives:

a2 2 a
∆i = (V in + V LO
2
) + a1Vin sin(ω in t) + a1VLO sin(ω LO t) − 2 V in2 cos(2ω in t) +
2 2
a 2
+ a2VinVLO cos(ω in − ω LO)t − V2VinVLO cos(ω in − ω LO)t − 2 V LO cos(2ω LO) + . . . (2.55)
2

where terms involving both ωin and ωLO are called intermodulation products. Provided that
ω in and ω LO are sufficiently close in frequency, the intermediate frequency, ω IF = | ω in − ω LO |,
is small compared to either the input signal frequencies or the local oscillator frequency and,
therefore, can be easily selected by filtering. A possible schematic diagram of a mixer circuit
is shown in Figure 2.29.

LPF ω IF
ωIF Output

BPF BPF
Schottky diode
ω In ω LO

ω In ω LO
Input Output

Figure 2.29 Schottky diode used in a mixer implementation


40 Microwave Devices, Circuits and Subsystems for Communications Engineering

2.5 PIN Diodes


A PIN diode is obtained by connecting a highly doped P + layer of semiconductor, a long
intrinsic (I) layer and a highly doped N+ layer. (Although, ideally, the I layer is intrinsic, in
practice the presence of impurities is unavoidable, making it slightly P or N doped.)
The presence of a long intrinsic section increases the breakdown voltage of the device thus
allowing high reverse voltages. This is advantageous when handling high input power. The
intrinsic section is also responsible for an almost constant value of reverse bias capacitance,
which is also comparatively smaller than that for P-N junctions. On the other hand, the
intrinsic semiconductor exhibits a variable resistance as a function of forward bias. Many
applications of the PIN diodes stem directly from this property, ‘PINs’ being often used as
variable resistances or variable attenuators. For a relatively high value of forward bias, the
resistance of the intrinsic section is noticeably reduced. Switching applications, between
reverse biased (no conduction or high resistance) state and forward biased (good conduction
or low resistance) state, are also possible.

2.5.1 Thermal Equilibrium


In thermal equilibrium the P + and N+ regions will, respectively, diffuse holes and electrons
into the intrinsic region, leaving ionised impurities. In each of these two regions the space
charge zone will have a reduced extension, since both are highly doped. Since the intrinsic
component is not doped, its corresponding density of net, static, charge is equal to zero. The
charge density profile, therefore, develops as shown in Figure 2.30(a).
The electric field variation along the device, obtained by integrating Poisson’s equation,
is shown in Figure 2.30(b). As can be seen, the electric field, E, takes a constant value along
the intrinsic part of the device, due to its zero static charge density.
The potential, V, is obtained by integrating the electric field. A linear variation is obtained
in the I region (Figure 2.30(c)), the potential is parabolic between −xp and 0 and between wI
and wI + xn, and elsewhere the potential is constant. The energy band diagram is shown in
Figure 2.30(d). The resulting potential barrier is higher than in P-N junctions.

2.5.2 Reverse Bias


When the PIN diode is reverse biased, there will be only a small saturation current due to
thermal electron-hole generation in the depleted zones, with the electrons moving towards
the N+ region and the holes towards the P + region.
In agreement with the principle of electric neutrality, the space charge zone is very narrow
in the P + and N+ regions due to the high doping density, while the intrinsic section is entirely
depleted. Thus, for any reverse bias it is possible to make the approximation wd ≅ wI, the
junction capacitance being given by:

ε sc S
Cj = (2.56)
wI

Unlike the case of a P-N junction, the above capacitance is almost independent of the bias
voltage and, due to the large width of the intrinsic region wI, has a small value. This gives
RF Devices: Characteristics and Modelling 41

ρ
ND+

0
(a)
–NA+
E
–Xp 0 WI WI + Xn
(b)
X

V V

(c) X
E WI

(d) EF

Figure 2.30 PIN diode charge density distribution (a), electric field distribution (b),
potential distribution (c), and energy band diagram (d)

rise to a very high reverse impedance, useful for switching applications. The avalanche
breakdown voltage, Vb, is given by:

Vb = EcwI (2.57)
where Ec is the critical field strength. Due to the large wI, the breakdown voltage is high,
so the PIN diode can utilise high reverse bias voltages. The critical field strength for silicon
is Ec = 2 × 105 V. Therefore, for a diode with wI = 10 µm the breakdown voltage is 200 V
while for a diode with wI = 50 µm the breakdown voltage is 1 kV.

2.5.3 Forward Bias


In forward bias carriers diffuse into the intrinsic zone from both P + and N− sides, resulting
in a forward current due to recombination of these ‘injected’ carriers. For a low bias voltage
the recombination takes place in the intrinsic region, while for a high bias voltage, it takes
place in P + and N+ regions.
Ignoring the intrinsic region doping, its carrier concentration will be mainly due to
electron and hole injection through the junctions. Assuming an equal concentration of
injected electrons and holes, p = n, the total drift current will be given by:
J = ( µ n + µ p)enE (2.58)
42 Microwave Devices, Circuits and Subsystems for Communications Engineering

the conductivity being:

σ = ( µ n + µ p)en (2.59)

Alternatively, the current flow, If , due to electron-hole recombination in the I region can be
calculated from:

Q
If = (2.60)
τ

where Q is the total injected charge and τ the average carrier life-time. The former is
given by:

Q = enSwI (2.61)

(Note that the electron and hole, moving in opposite directions before recombining, count as
a single charge in the recombination current.)
The resistance of the intrinsic zone can be calculated from its conductivity value. Assuming
µ p ≅ µ n ≅ µ, the resistivity, ρ, will be:

1
ρ= (2.62)
2 µen

and the resistance, RI, of the intrinsic zone is therefore given by:

ρwI wI
RI = = (2.63)
S 2 µenS

From Equation (2.61):

Q
en = (2.64)
SwI

And substituting Equation (2.64) into Equation (2.63):

SwI wI w2
RI = = I (2.65)
2 µQ S 2 µQ

Expressing the charge as a function of the forward current, Q = If τ, a relationship is obtained


between the intrinsic resistance and this current, i.e.:

wI2
RI = (2.66)
2 µτ If

The intrinsic resistance, RI, is therefore inversely proportional to the forward bias current,
If , and it is this relationship which enables the diode to be used as a variable resistor.
RF Devices: Characteristics and Modelling 43

Since the intrinsic zone is not doped, its conductivity is notably smaller than that of the P +
and N+ regions resulting in a capacitance, CI, [2]:

ε sc S
CI = (2.67)
wI − wdI

where wdI is the extension of the depleted zone in the I region. Note that this capacitance is
exclusively due to the intrinsic material, so it vanishes if the I region shrinks to zero width.
(For reverse bias wdI = wI and the capacitance CI tends to infinity – its effect therefore
disappearing in the context of overall device performance. This is more clearly understood
with the aid of the circuit electric model given in Section 2.5.4 below.)
In forward bias, the I region stores a charge Q composed of charge carriers. If the diode
state is abruptly switched to reverse bias, a non-zero time, τs, is required to purge the I
region of these charges. This is the diode switching time which will be higher for longer
lifetime, τ, and larger intrinsic region width, wI. (When switching from reverse to forward
bias, the time needed for carrier injection is notably smaller.)

2.5.4 Equivalent Circuit


The equivalent circuit for a PIN diode (Figure 2.31) is, principally, composed of two parallel
circuits in series combination, one circuit accounting for the two P-N junctions and the other
for intrinsic section.

Lp

Rs

+
Cj V I(V) Cd
Cp

RI CI

Figure 2.31 Equivalent circuit model for PIN diode


44 Microwave Devices, Circuits and Subsystems for Communications Engineering

The former contains a non-linear current source in parallel with both junction and diffu-
sion capacitances. (Note that both junctions are accounted for by these capacitors.) The latter
contains a variable resistance, Ri, in parallel with the intrinsic capacitance Ci. A constant loss
resistance, Rs, is also usually present, accounting for the metallic contact resistance and the
resistance of the neutral zones P and N (which have a limited conductivity value).
In addition to the above equivalent circuit elements, packaging parasitics may be included.
The equivalent circuit for the packaging comprises a parallel capacitor, Cp, due to the capacit-
ance between the device contacts, and a series inductor, Lp, due to the bonding wires.
In reverse bias, the part of the equivalent circuit representing the junctions is reduced to
the junction capacitance, Cj, alone since the diffusion capacitance, Cd, vanishes for negative
bias and the current source supplies negligible current. The equivalent circuit representing
the intrinsic section also disappears since, for reverse bias, the whole intrinsic section becomes
ionised, behaving as a depleted region. As stated in Section 2.5.3 above, the capacitance CI
tends to infinity and short-circuits the resistance RI.
In forward bias, the diffusion capacitance short-circuits the non-linear current source.
The diode equivalent circuit is then reduced to the intrinsic component of the model, i.e. the
variable resistor RI and the parallel capacitor CI. The value of RI decreases rapidly with
increasing forward bias, so for relatively high bias voltages, the equivalent circuit reduces to
a small resistor. This is well suited to switching applications, the high-impedance state being
provided by reverse bias.

2.5.5 Manufacturing
PIN diodes are manufactured in mesa technology. The substrate is highly doped silicon N+.
One of the terminals is fabricated at the device base, as an ohmic contact between the N+
substrate and a metal layer (Figure 2.32).
On the upper surface of the substrate a non-doped layer constitutes the intrinsic region.
Due to manufacturing imperfection, this will be either slightly P or N doped in practice.
Acceptor impurities are diffused, or implanted, over the surface of the intrinsic layer to
realise the P+ region. The second device terminal is fabricated as an ohmic contact between
the P+ layer and a metal deposit.

Sup. S
P+

d I

N+

Figure 2.32 PIN diode structure


RF Devices: Characteristics and Modelling 45

Self-assessment Problems
2.13 Why does the intrinsic part of resistance in a PIN diode quickly decrease with
forward bias?

2.14 Why does the reverse capacitance of a PIN diode remain approximately constant
with variations in reverse bias voltage?

2.5.6 Applications
The special applications of PIN diodes are made possible by the long intrinsic region
embedded between the P + and N+ zones. This increases the breakdown voltage enabling PIN
devices to deal with high input powers. Small reverse bias capacitance and low forward bias
resistance make PIN devices useful for switching applications. The variable resistance property
in forward bias also enables PIN devices to be used as variable attenuators.

2.5.6.1 Switching
In switching applications, advantage is taken of the PIN diode’s low, and almost constant,
reverse bias capacitance, Cj, providing very high input impedance. A low-impedance state
is obtained by forward-biasing the diode, since the resistance RI decreases rapidly with the
applied voltage. The high breakdown voltage in these devices allows the switching of high
power signals. An important PIN diode characteristic that must usually be considered in
switching applications is the switching time, τs. Switching time is determined by carrier
lifetime, shorter switching times being obtained for shorter carrier lifetimes.
In the design of switching circuits, different topologies are possible. Two of these topologies
are analysed below.

Topology 1
A simple switch [4] may be realised by connecting a PIN diode in parallel over a transmis-
sion line (Figure 2.33).
In the low-impedance state, corresponding to forward bias, there will be high reflection losses
and only a small proportion of the input power will reach the load. In the high-impedance
state, corresponding to reverse bias, the diode will give rise to only a small insertion loss.

Zc
Zc
Eg

Figure 2.33 Switch realised using a PIN diode and transmission line
46 Microwave Devices, Circuits and Subsystems for Communications Engineering

For a quantitative analysis of this switch, the attenuation corresponding to both the forward
and reverse states must be calculated. Let Y be the diode admittance and Z0 the characteristic
impedance of the transmission line. In order to derive a simple expression for the switch
attenuation, the connection of the general admittance, Y, across the transmission line is going
to be considered. The input generator has a voltage Eg and internal impedance Z0 and the
load impedance is matched to the characteristic impedance of the line, i.e. ZL = Z0. The
power delivered to the load is given by:

1 Eg2
PL = (2.68)
Z0 2 | 2 + y |2
where y is the normalised admittance, y = YZ0. If the admittance Y were not connected, then
all of the generator’s available power would be delivered to the load, i.e.:

Eg2
Pdg = (2.69)
8Z0
The attenuation, A, due to the presence of admittance, Y, is therefore given by the ratio of
Equations (2.69) and (2.68), i.e.:

Pdg | 2 + y |2
A= = (2.70)
PL 4
The forward-biased normalised PIN admittance, yF, is dominated by the intrinsic resistance,
RI, plus the loss resistance, RS. Thus:

Z0
yF = (2.71)
RI + RS
The reverse-biased, normalised, PIN admittance, yR, is given by the junction capacitance
alone. Therefore:
yR = jZ0Cj ω (2.72)
The attenuation for forward- and reverse-biased PIN states is then found by substituting the
corresponding normalised admittance into Equation (2.70).

Topology 2
In Figure 2.34, the input RF signal may be transferred to channel A or channel B, depending
on the state of the corresponding PIN diodes [4].
The two transmission lines (of length λ /4) behave as impedance inverters. The inductors
and capacitors in the schematic behave, respectively, as DC-feeds and DC-blocks.
Consider diode DA in forward bias and diode DB in reverse bias. Due to the λ /4 line,
channel A will exhibit a very high impedance to the input signal. Since DB is reverse-
biased, its impedance will be high and, connected in parallel across the line, it will have little
effect on transmission. The entire input signal will therefore be delivered to channel B.
Conversely, when DA is reverse biased and DB is forward-biased, the entire input signal is
delivered to channel A.
RF Devices: Characteristics and Modelling 47

LDC λ /4 Cb Cb λ /4 LDC

A B
DA DB
CHANNEL CHANNEL

Inp
ut
Figure 2.34 Double throw switch implemented by transmission lines and PIN diodes

2.5.6.2 Phase shifting


The phase shifting of high power signals may be obtained with PIN diode circuits. Several
circuit topologies are possible, two of which are analysed below.

Topology 1
The topology of Figure 2.35 is based on the transfer characteristics of a transmission line,
loaded at both ends by equal susceptance, jB.
The phase shift suffered by a microwave signal traversing the complete two-port can
be found by calculating the transmission matrix of the two-port and transforming it into a
scattering matrix – see Chapter 3 and [3]. The phase of the S21 element of the scattering
matrix directly provides the phase shift suffered by the signal. For the particular case of a
quarter wave line (l = λ /4), this phase shift is given by:

 2 − B2 Z 02 
ϕ = tan −1   (2.73)
 2 BZ0 

where Z0 is the characteristic impedance of the transmission line. It is clear from Equation
(2.73) that modifying the susceptance, B, varies the phase shift.

3 3
L L
2 2
Input jB jB Ouput
1 1

Figure 2.35 High power phase shifter implemented using transmission line and PIN diodes
48 Microwave Devices, Circuits and Subsystems for Communications Engineering

When a PIN diode with negligible parasitics is connected at each end of the transmission
line, two different phase-shift values can be obtained. These respectively correspond to the
forward-bias state, with BF = 0, and the reverse-bias state, with BR = Cjω. In order to increase
the differential phase shift between the two states, each susceptance B may be implemented
using a PIN diode in series with an inductor L [4]. The inductance is chosen to satisfy:

1
Lω = (2.74)
2 C jω
With the diode in reverse bias the total reactance is −1/(2Cjω) and the corresponding
susceptance BR = 2Cjω. With the diode in forward bias the susceptance is BF = −2Cj ω.

Example 2.3
A PIN diode is connected at each end of a 50 Ω quarter-wave transmission line to realise a
microwave phase shifter. Each diode has a reverse bias capacitance (Cj) of 0.175 pF and the
required operational frequency of the phase shifter is 5 GHz. Find the differential phase shift
with, and without, an appropriate series inductor connected to the diodes.
When the diodes are reverse-biased, the susceptance value is BR = 5.5 10−3 S and the
corresponding phase shift ϕR = 74°. When the diodes are forward-biased, the susceptance
value is BF = 0 and the corresponding phase shift ϕF = 90°. The differential phase shift is
∆ϕ = 16°. When including a series inductor, calculated according to Equation (2.73), the
phase shift in the forward bias state is ϕ F = 57°. In the reverse bias state, the phase shift is
ϕR = −57°. The differential phase-shift is ∆ϕ = 114°.
The principal problem with the above topology is input and output mismatching. The
parallel susceptance varies the input and output impedances of the network and this may
dramatically increase the reflection losses. In order to avoid this drawback, a different
topology (described below), based on the use of a circulator, can be employed.

Topology 2
In the topology of Figure 2.36, assuming an ideal circulator, power incident on port 1 is
entirely transferred to port 2. The power reflected by the load of port 2 is transferred to port
3 and thence to the circuit-matched load.

lx l l l
Zg
1 2 D1 D2 D3 DN
Eg
3

ZL

Figure 2.36 High power phase shifter implemented using transmission line,
circulator and PIN diodes
RF Devices: Characteristics and Modelling 49

A number of PIN diodes are connected in parallel across the transmission line connected
to port 2. The first diode is located at a small distance lx from the circulator [3, 4]. The distance
between any other two diodes in the array is constant and equal to l. When the first diode,
D1, is forward-biased, the input-output phase shift is given by:
φ1 = π − 2βlx (2.75)

where β is the phase constant (in rad/m) of the line. When all the diodes up to Dn are reverse-
biased and Dn+1 is forward-biased, the phase shift is given by:
φn+1 = π − 2βlx − 2βnl (2.76)

This topology generally provides a larger phase-variation range than topology 1. Its main
advantage is the input and output impedance matching provided by the circulator.

Example 2.4
Design a phase-shifter from 0° to 180° in steps of 45°, using PIN diodes. The operating
frequency is 4 GHz and the available substrate has an effective dielectric constant εeff = 1.86.
Since five discrete phase shifts are required, five diodes are needed and the maximum
value of n in Equation (2.76) is 4. Ignoring lx, when the first four diodes are reverse biased
(off) and diode D5 is the first to be forward biased (on), the phase shift is given by:

φ5 = π − 8βl
The phase shift φ5 is made equal to 0°, so the length of the transmission line sections must
have the value:
π
l=

The phase shift φ1 = π will be given by the state D2 to D5 off and D1 on. The rest of phase
shift values are given by:
π
φn +1 = π − n
4
where n is the index of the last diode in off-state.
In order to complete the design, the physical length of the transmission line must be
calculated. At f = 4 GHz the wavelength value is:

c
λ= = 0.055 m
f ε eff
and:

β= = 114.25 m −1
λ
Therefore:
l = 3.44 mm
50 Microwave Devices, Circuits and Subsystems for Communications Engineering

Zc C
Bias
Zc
Eg

Figure 2.37 PIN-based attenuator

2.5.6.3 Variable attenuation


The variation of the intrinsic resistance, RI, of the PIN diode as a function of bias current, If,
allows its application as the active device in a variable attenuator. The resistance/current
relationship is:
α
RI = (2.77)
If
where the parameter α is given by:
wI2
α = (2.78)
2τµ
and wI, τ and µ are intrinsic region width, average carrier lifetime and average carrier
mobility respectively. A simple implementation of a PIN-based attenuator circuit is shown
in Figure 2.37.
The PIN diode with a series (bias blocking) capacitor is connected across a transmission
line. When the bias current, If , is modified, the corresponding normalised admittance y varies
according to:
Z0
y( I f ) = (2.79)
α
+ RS
If

that results in an attenuation (Equation (2.70)) of:

| 2 + y( If ) |2
A( I f ) = (2.80)
4

2.5.6.4 Power limiting


For power limiting applications, the PIN diode is connected in parallel with the receiver or
device (e.g. detector, mixer, amplifier) to be protected as shown in Figure 2.38.
The inductor, L, enables the DC current circulation. The diode eliminates or attenuates all
the signals with amplitude higher than a certain threshold [3, 4, 5].
For small signals, the diode operates around 0 V, exhibiting high impedance, due to the
small junction capacitance characteristic of PIN diodes. In this case the signal transmitted to
the receiver is almost unperturbed. For large signals (i.e. when the input amplitude is high),
RF Devices: Characteristics and Modelling 51

Input
L Receiver
Signal

Figure 2.38 PIN-based power limiting for receiver protection

the diode impedance greatly decreases at the signal (positive going) peaks, leading to deep
forward bias. For high operating frequencies, the injected carriers are only partially evacu-
ated during the negative going (reverse bias) peaks. Since the diode is still conducting due to
the residual carriers, its impedance remains low for the entire cycle of the input waveform.
In this way the signal delivered to the receiver is greatly attenuated.
Higher attenuation values are obtained for smaller intrinsic region widths, due to the
larger junction capacitance. When handling high input powers, however, the intrinsic region
must be sufficiently wide to avoid avalanche breakdown.

2.6 Step-Recovery Diodes


Step-recovery diodes (also called snap-off diodes) are based on a PIN configuration. They are
commonly employed in the design of frequency multipliers of high order. Their operating
principle is as follows.
When a PIN diode is forward-biased, the two junctions P +I and IN+ inject carriers into the
I region. The recombination of these carriers after a time τ gives rise to the forward current
If. If the diode is now reverse-biased through the application of a voltage step, the injected
carriers that are present in the I region will have to be removed. There will be a large reverse
diffusion current limited only by the external circuit. The carrier concentration, however,
does not decrease uniformly along the I region. There will be a faster decrease close to the
junctions. After a time, ta, the concentration level at the junctions becomes zero. However,
there will still be stored charge in the I region that is removed, as in a capacitor, with a
certain time constant tb. It is possible to artificially increase ta, using a special doping
process. The aim is to increase ta reducing the number of carriers still stored in the I region
after this time. The smaller number of stored charges gives rise to a much smaller time tb.
The decrease in tb compared to ta provides a pulsed response. When using a sinusoidal input
voltage, this pulsed current can be used for harmonic generation. It is thus possible to obtain
frequency multiplication of high order.
In the design of high-order frequency multipliers, the efficiency of step-recovery diodes is
much higher than that of varactor diodes. There is, however, a limit to the output frequency
of the multiplier circuit. The total switching time, ta + tb, must be much smaller than the
reciprocal of the output frequency for proper multiplier operation. If this is not the case, very
high conversion loss is obtained. A common criterion for the estimation of losses is based
on the calculation of the threshold frequency ω th = 1.6/(ta + tb). For output frequencies below
ω th the conversion loss increases at approximately 3 dB/octave. Above this frequency it
increases at approximately 9 dB/octave.
52 Microwave Devices, Circuits and Subsystems for Communications Engineering

Self-assessment Problems
2.15 What is the phenomenon giving rise to the high reverse current that is initially
observed when switching a PIN diode from forward to reverse bias?

2.16 What limits the output frequency of step-recovery diodes used for frequency
multiplication?

2.7 Gunn Diodes


Gunn diodes are manufactured using III–V compound semiconductors, such as GaAs or PIn
(Phosphorus-Indium). They exhibit a negative resistance, related to the multi-valley nature
of their conduction bands. The oscillating properties of these materials were discovered (by
Gunn), in 1963, during his studies of GaAs noise characteristics. (Gunn was working on a
biased GaAs sample when he detected RF oscillations in the microwave range.)
Gunn diodes are usually fabricated in GaAs. In the conduction band of this material,
a main valley, 1.45 eV above the valence band, is surrounded by six satellite valleys with
0.36 eV higher energy (Figure 2.39).
Electrons residing in the main valley have lower energy and higher mobility compared
with electrons in the satellite valleys.
Gunn diodes are manufactured in three layers, with an N-doped layer embedded between
two, thinner, N+-doped layers (Figure 2.40).
The two external layers facilitate the ohmic contacts that provide the device terminals.
Consider a Gunn diode connected to an external bias source. As the bias voltage is increased
from a low value, the electrons acquire a higher energy and their velocities increase. When
the field strength reaches a threshold, Eth, electron collisions become sufficiently frequent
(due to their enhanced velocities) that their predilection to drift in the electric field becomes

m 2*
m 1* η2
COND.
η1 0.36 eV

1.45 eV
AsGa – N

VAL

Figure 2.39 Energy band diagram for GaAS


RF Devices: Characteristics and Modelling 53

V
− +

N+ N N+

Figure 2.40 Three-layer structure of Gunn diode

significantly impaired; in short, their mobility is reduced. This corresponds to electrons in


the lower energy valley being transferred to the higher energy valleys.
The relative concentrations of electrons in the lower and higher energy valleys depend on
the electric field strength. These concentrations are denoted by n1 (E) for the lower energy
valley and n2 (E) for (all) the surrounding valleys of higher energy. Three different ranges of
electric field strength may be considered [1, 2, 3]:

(a) For 0 ≤ E ≤ Eth all electrons occupy the lower energy valley, so the total electron con-
centration n = n1. The electron velocity, in this regime, increases linearly with increasing
electric field, i.e. v = −µ1 E. This corresponds to the electrons in the lower energy valley
acquiring more kinetic energy. The current density is, therefore, given by:

J = µ 1en1 E (2.81)

(b) For Eth ≤ E ≤ E2 some of the electrons in the lower energy valley acquire sufficient
energy to reach the higher valleys. The total number of electrons is then distributed
between the low and high energy valleys, i.e. n(E) = n1 (E) + n2 (E). The higher energy
valley has a lower mobility, µ2, due to the collision effect described above and the current
density is, therefore, given by:

J = ( µ 1n1 + µ 2 n2)E (2.82)

The ratio n2(E)/n(E) increases with increasing electric field. The current density, however,
decreases since the reduction in average electron mobility (due to the increasing proportion
of slower electrons) outweighs the effect of the increased electric force on each electron.

(c) For E ≥ E2 all electrons occupy the higher energy valleys, so n = n2. Since there is now
no transference of electrons from lower to higher energy valleys, electron velocity will
again increase with applied field, i.e. v = −µ2 E. The current density is now given by:

J = µ2 en2 E (2.83)

Figure 2.41 shows mean electron velocity plotted against applied electric field strength [1].
The behaviour of velocity in this figure arises due to the non-linear variation of mobility
as a function of the applied field, i.e. v = −µ(E)E. This results in a negative differential
54 Microwave Devices, Circuits and Subsystems for Communications Engineering

ϑ ·107 cm/s n2 /n1 + n2

2 1

1.5 0.75

1 0.5

0.5 0.25

0 2 Et 4 6 8 10 E (kv/cm)

Figure 2.41 Mean electron velocity versus applied electric field

mobility in a certain electric field range and thus in negative resistance. Note that current
depends on the carrier velocity, while voltage drop is related to the electric field. A plot of
the ratio n2 (E)/n(E) is included in Figure 2.41 showing the close relationship it has with
velocity variation. Negative-resistance is only observed if the period is shorter than the time
required to transfer electrons from the low-energy to the high-energy valley. The electric
field threshold needed to obtain negative differential mobility therefore varies with operating
frequency.

2.7.1 Self-Oscillations
Provided the electric field remains below the threshold value Eth, the electric potential
decreases smoothly along the device (Figure 2.42(a)).
For electric fields above the threshold, however, some electrons will be slowed down due
to their transference to the higher energy valleys. The slow electrons will bunch with the

V V
− + − +

− +
C A C A

(a) (b)

Figure 2.42 Electric potential versus position along Gunn diode: (a) electric field below threshold,
(b) electric field above threshold
RF Devices: Characteristics and Modelling 55

It

Figure 2.43 Gunn current versus time

following faster electrons giving rise to a negative space charge accumulation [4]. (The
precise location where the space charge initially forms is determined by random fluctuations
in electron velocity or by a permanent non-uniformity in the doping.) Every delayed electron
gives rise to a negative charge deficiency in front of, and an excess negative charge behind,
its location had it not been delayed. This effect is cumulative and the negative space charge
accumulation is compensated for by a positive space charge of equal magnitude. The result
is a space charge dipole that moves from cathode to anode at the electron-saturated velocity
(Figure 2.42(b)). The potential now drops unevenly along the diode, the gradient being
greatest across the dipole (Figure 2.42(b)) and the electric field along the rest of the device
necessarily remaining below the threshold value. The reduced value of electric field away
from the space charge dipole prevents more than one dipole forming at any one time.
When the slowly moving dipole reaches the anode, a current peak is detected (Figure 2.43).
The old dipole is now extinguished and a new one will immediately appear, since the
electric field in the device is again above the threshold value. This will reduce the detected
current. In this way, by simply biasing the semiconductor, it is possible to obtain a non-
sinusoidal oscillation of current at the frequency ft = 1/τ where τ is transit time across the
device of the space charge dipole. (The transit time of the space charge dipole is, of course,
equal to the transit time of individual electrons.)
It is sometimes important to appreciate that the load circuit connected to a Gunn diode
may reduce the electric field to values below the threshold value. If this is the case, then
the space charge dipole will disappear during one part of the high frequency cycle. As will
be shown in the following, the load circuit connected to the Gunn device has great influence
on the device operating mode.

2.7.2 Operating Modes


Gunn devices can operate in different modes [1], depending on factors such as doping
concentration, doping uniformity, length of the active region, type of load circuit and bias
range. The formation of a strong space charge instability (i.e. space charge dipole) requires
enough charge, and sufficient device length, to allow the necessary charge displacement
within the electron transit time.
56 Microwave Devices, Circuits and Subsystems for Communications Engineering

The boundary between the different modes of operation is denoted by the product n0 l,
where n0 is boundary carrier concentration for a given device length, l [1]. For GaAs and InP
devices, the boundary condition between stable field distribution and space charge instability
is given by n0 l = 1012 cm−2.

2.7.2.1 Accumulation layer mode


A Gunn device with sub-critical nl product (nl < 1012 cm−2) is now considered. An accumula-
tion layer starts at the cathode, propagates towards the anode and disappears at the anode.
The field at the cathode then rises, enabling the formation of another accumulation layer,
the process thereafter repeating indefinitely. This oscillation mode has a low efficiency, but
advantage can be taken from the negative resistance exhibited by the device in a frequency
band near the reciprocal of the electron transit time and its harmonics. In this mode the
device can therefore be operated as a stable amplifier. The frequency of the peak gain of
such an amplifier increases with applied electric field strength. Measurements show that
negative resistance can be obtained over a frequency band greater than one octave.

2.7.2.2 Transit-time dipole layer mode


Transit-time dipole layer operation is obtained for supercritical values of the product nl.
Dipole layers form and propagate to the anode. The cyclic formation of these dipoles and
subsequent disappearance at the anode constitute the self-oscillations. The efficiency, how-
ever, is only a few percentage. To be able to use dipole transit for self-oscillation the quality
factor of the load circuit must be low to avoid a reduction in the electric field during the
negative part of the cycle. It is this that depresses efficiency and keeps interest in this
operating mode low.
The expected frequency of the self-oscillation may be found from vs = l/τ, where vs the
space charge (saturated) velocity and l the device length. A filter will be necessary in order
to eliminate the higher order harmonic components if sinusoidal oscillation is required.
Typically the DC bias voltage is three times the threshold voltage, Vth = Eth l.

2.7.2.3 Quenched dipole layer mode


In devices with a supercritical nl product, it is possible to obtain oscillation frequencies
higher than the reciprocal of the transit-time frequency by quenching the dipole layer using
a resonant circuit. The oscillating voltage from the resonant circuit adds to the bias voltage
reducing the total voltage during its negative half cycles. The width of the dipole layer is
reduced during what is, effectively, these periods of reduced bias. Dipole layer quenching
occurs when the bias voltage is reduced below the threshold value Eth. When the effective
bias voltage increases above this threshold (during the positive half cycles of the oscillating
voltage), a new dipole layer forms, the process then repeating indefinitely. The oscillations
therefore occur at the frequency of the resonant circuit rather than the reciprocal of the
transit time frequency.
If the quality factor of the load circuit is high, the amplitude at the diode terminals may be
as high as the bias voltage. During sign opposition the electric field will decrease to a very
low value and the dipoles will disappear during part of the cycle. If dipoles disappear before
RF Devices: Characteristics and Modelling 57

reaching the anode, the oscillation frequency will be higher than the transit frequency ft . If
the electric field is low, the dipole may traverse the whole device length l, with the appearance
of a new dipole being delayed. This would give rise to an oscillation frequency smaller
than ft. The efficiency in this oscillation mode is much higher than the one obtained in the
transit-time dipole-layer mode.

2.7.2.4 Limited-space-charge accumulation (LSA) mode


In limited-space-charge mode, the dynamic operating point is located in the regions of positive
resistance for most of the oscillation period. The LSA mode assumes large dynamic swings
around the operating point. A high frequency, large amplitude, voltage reduces the electric
field during the cycle to a value slightly lower than Ec that suppresses the formation of dipoles.
The frequency must be high enough to prevent the dipole formation (i.e. there must not
be enough time for dipole formation). The electric field across the device rises above, and
falls back below, the threshold so quickly that the space charge distribution does not have
sufficient time to form. The mean resistance over one cycle, however, must be negative and
to fulfil this condition, the electric field must not fall to values much lower than Ec, which
requires a limited space charge. The diode behaves like a negative-real-part impedance
without producing pulses. Oscillation frequency is determined by the external circuit.
LSA mode operation yields much higher efficiency than that obtained from dipole transit.
Furthermore, there is no transit time limitation on the maximum oscillation frequency.

2.7.3 Equivalent Circuit


The most commonly used equivalent circuit for a Gunn diode consists of non-linear current
source with an N-shaped current-voltage characteristic, a capacitance and a loss resistance.
Series and parallel circuit configurations are possible, the choice depending on bias point,
bandwidth and operating frequency. A series configuration is shown in Figure 2.44.
For small signals, and provided that the diode is biased in the negative slope region, the
non-linear current source may be replaced by a negative resistance.
In order to implement ohmic contacts, the semiconductor regions close to the metallic
terminals are highly doped and thus more conductive than the central region. The capacitor
C accounts for the capacitance between these contacts. The loss resistance r arises due to the
limited conductivity of the substrate.

L C r

I(V)

Figure 2.44 Series equivalent circuit model for a Gunn diode incorporating non-linear current source
58 Microwave Devices, Circuits and Subsystems for Communications Engineering

Self-assessment Problems
2.17 What kind of materials can exhibit the Gunn effect?

2.18 What is the phenomenon that gives rise to the dipole layer formation in the Gunn
diode? Why is only one dipole layer observed along the entire device length?

2.19 How does load circuit quality factor influence the operating mode of the Gunn
diode?

2.7.4 Applications
Gunn devices are used, principally, in the design of negative resistance amplifiers and
oscillators.

2.7.4.1 Negative resistance amplifiers


Consider a Gunn diode, biased with a certain DC voltage. The series equivalent circuit
(Figure 2.45) includes a negative resistance, −Rn < 0, in series with a capacitor, Cd, and a
loss resistance, r.
The negative resistance −Rn is defined by:
−1
 dI 
− Rn =   (2.84)
 dV V0 

where I(V) is the non-linear current source characteristic and V0 the device bias voltage.
The net negative resistance is −Rd = −Rn + r, with Rd > 0 implying a voltage reflection
coefficient, Γ, at the device terminals with an amplitude larger than one (i.e. | Γ | > 1).

Cd
L

–Rd(I)
RL

Load
Diode

Figure 2.45 Series equivalent circuit of Gunn diode with DC bias and negative resistance
RF Devices: Characteristics and Modelling 59

(This implies a reflected power that is higher than the incident power, and thus genuine
amplification.) Depending on the value of the passive load, different gains and bandwidths
can be obtained. A particularly simple amplifier implementation is realized when incident
and reflected powers are separated using a circulator.

2.7.4.2 Oscillators
Gunn diodes are often employed as the active device in microwave oscillators. In the
dipole layer mode, device length determines oscillation frequency. Around this frequency,
the diode exhibits negative resistance and, when using the accumulation layer mode, the
precise oscillation frequency may be fixed by the resonant frequency of an external (passive)
circuit. The main attractions of Gunn-based oscillators are their capability to operate over
a large frequency band, their low noise performance and their high output power. Gunn
oscillations have been obtained up to frequencies of about 150 GHz and output powers of
15 mW at 100 GHz can be realised. Their main disadvantage is low DC-RF efficiency.
Another disadvantage is their temperature sensitivity, an effect inherent in their working
principle.
A small signal model for the Gunn diode consists of a capacitance Cd and a total negative
resistance –Rd (including losses). A simple oscillating configuration is shown in Figure 2.47,
in which the diode load is composed of a resistor plus an inductor in series configuration.
For oscillation to start, the circuit total resistance (including linear and non-linear contributions)
must be negative at the resonance frequency, ω 0, given by:

1
ω0 = (2.85)
LCT
where:

CCd
CT = (2.86)
C + Cd

2.8 IMPATT Diodes


IMPATT (Impact Avalanche and Transit Time) diodes are very powerful microwave sources,
providing the highest (solid state device) output power in the millimetre-wave frequency
range. They exhibit a dynamic negative resistance based on transit time effects and they are
often used in the design of oscillators and amplifiers when high output power is required.
They are manufactured in Si, GaAs and InP.
IMPATT diode provides negative resistance using the phase shift between current through
the device and the applied voltage. (Negative resistance is obtained when this phase shift
is greater than 90°.) The device consists of a reverse-biased P-N junction (operating in
avalanche breakdown) and a drift zone. Carriers are injected into the drift zone from the
junction in avalanche breakdown where they drifted at saturated velocity. The constant
value of the saturated velocity provides a linear relationship between the device length and
the current delay. It is, therefore, possible to determine a length that will result in negative
resistance within a certain frequency band.
60 Microwave Devices, Circuits and Subsystems for Communications Engineering

Like Gunn diodes, IMPATT diodes are generally used in the design of negative resistance
amplifiers and oscillators. They provide higher output power than Gunn devices, and may
be operated at frequencies up to about 350 GHz when manufactured in silicon. They are
noisier than the Gunn devices, however, a characteristic inherent in their operating prin-
ciple. Due to their noisy operation they are seldom used for local oscillators in receivers.
Another drawback is the low value of their negative resistance that can give rise to matching
difficulties.

2.8.1 Doping Profiles


IMPATT diodes are manufactured with many different doping profiles. In all variations,
however, avalanche and drift regions can be distinguished [3]. Single drift profiles occur in
P+NN+ and N+PP+ structures, the former being preferentially manufactured, since N type
substrate is the more common. In P+NN+, the P-N junction is biased near avalanche break-
down. The P region injects electrons into the NN+ region, along which they drift at saturated
velocity. The holes injected from the N region do not drift. Hence the name of single drift
profile. Double drift devices provide a higher efficiency and a higher output power. One
possible double drift structure is P+PNN+. The central P-N junction operates in avalanche
breakdown, injecting electrons that drift along the NN+ region, and holes that drift along the
P+P region. Other possible profiles include the Read configurations. For single drift devices
Read proposed N+PIP+ and P+NIN+ structures. The former is analysed below.
Although IMPATT diodes are manufactured in both Si and GaAs, higher efficiencies are
obtained with the latter. Mesa technology is used for their fabrication. In the case of a P+NN+
profile, for example, two layers (N and P+) are diffused by a double epitaxial growth process
over a highly doped N+ layer. Ionic implantation is also possible. Finally, the ohmic contacts
are realised through a metal deposit.

2.8.2 Principle of Operation


A device with an N+PIP+ profile (Figure 2.46) is analysed here to illustrate the IMPATT’s
principle of operation. Assume a reverse bias V = Vb is applied to the IMPATT device, where
Vb is the breakdown voltage. A sinusoidal waveform is then superimposed, resulting in
a total device voltage v(t) = Vb + V1 cos ω t. When v(t) > Vb, a strong ionisation process
(breakdown) starts at the N+P junction resulting in the generation of electron-hole pairs.
The electrons move towards the positive terminal, while the holes drift at saturated velocity
vs along the depletion zone (i.e. intrinsic, or I zone) towards the P + region.
The drift time will be shorter for the electrons from P, due to the shorter distance to be
traversed. Their effect on the external current will not be observed until the holes reach P+,
however, in good agreement with the principle of electric neutrality. The device length may
be chosen to provide the necessary delay for a 180° phase-shift between the device voltage
and current.
The applied voltage, consisting of a sinusoidal waveform, superimposed on the bias
voltage Vb is illustrated in Figure 2.47(a).
As shown in Figure 2.47(b), as long as v(t) > Vb, the number of carriers increases,
even beyond the voltage maximum. This is explained by the fact that, during breakdown,
electron-hole generation depends on the total carrier number. Since the avalanche current is
RF Devices: Characteristics and Modelling 61

V
+ −

N+ P O I P+

Figure 2.46 Schematic structure of IMPATT diode

Vb (a)

(b)

(c)

T
2

Figure 2.47 Applied voltage (a), carrier density (b) and current for IMPATT diode

directly proportional to the carrier concentration, this current will have a one-quarter period
(T/4) delay with respect to the applied signal voltage. In order to obtain the desired 180°
phase-shift between applied voltage and device current, a further T/4 delay is necessary.
This is provided by the hole drift along the depletion region. Since the holes move at the
constant velocity, vs, the necessary device length is given by:

T
l = vs (2.87)
4

Example 2.5
The carrier saturation velocity in silicon semiconductors is vs = 105 m/s. What should be the
length of a silicon IMPATT in order to obtain a negative resistance at about 12 GHz?
62 Microwave Devices, Circuits and Subsystems for Communications Engineering

In order to obtain a negative resistance, the carrier drifting through the device must give rise
to a phase shift of π /4, to be added to the avalanche phase shift of π /4. The drift time must
therefore be T/4. Taking Equation (2.85) into account:

T
l = vs
4
where:
1
T= = 8.33 × 10−11 s
12 × 10 9
Therefore:
l = 2 µm

2.8.3 Device Equations


Ionisation rate is the probability of an electron-hole generation per carrier unit length. It can
be calculated from the following formula [3]:

  b  m
α = C exp  −  (2.88)
  E 

where m, b and C are parameters which depend on the material and type of particle.
Electrons and holes therefore have different ionisation rates, α n and α p.
Since electric field is, in general, a function of distance x along the device, so are the
ionisation rates. The electric field dependence can be clarified by remembering Poisson’s
equation, i.e.:
∂E e
= [ N D − N A + p − n] (2.89)
∂x εs
Integrating Equation (2.89) gives the electric field variation, E(x), along the device and know-
ing this the ionisation rate, α, as a function of the x can be determined from Equation (2.88).
It can be shown that the condition for avalanche breakdown is given by [1]:

w  x 
 
冮 冮
α p exp − (α p − α n )dx ′ dx = 1
 
0
 0 
(2.90)
w  w 
 
冮 冮
α n exp − (α n − α p )dx ′  dx = 1
 
0
 x 

where w is the width of the avalanche region.


RF Devices: Characteristics and Modelling 63

Recall the carrier continuation Equations (2.20) and (2.21). For generation by impact
avalanche, the coefficients, gp and gn, are given by:

α n Jn + α p Jp
gp =
e
(2.91)
α J + α p Jp
gn = n n
e

The continuity equations for electrons and holes are given by:

∂n 1  ∂Jn 
=  + α n Jn + α p Jp 
∂t e  ∂x 
(2.92)
∂p 1  ∂J p 
=  + α n Jn + α p Jp 
∂t e  ∂x 

where diffusion current in the drift region has been neglected.


The total current density J is:

J = Jn + Jp (2.93)

J may also be written:

∂E
J = envn + epvp + ε s (2.94)
∂t

where vn and vp are, respectively, the electron and hole-saturated velocities.


Equations (2.90), (2.92), (2.93) and (2.94) comprise a system of five equations (Equa-
tions (2.93) and (2.94) counting as one equation) in five unknowns (E, p, n, Jn and Jp) that
may be solved numerically [3]. Once E and J are obtained the non-linear current-voltage
relationship of the IMPATT diode can be found using:


V = Edx
x
(2.95)

2.8.4 Equivalent Circuit


In order to obtain a small signal model for the IMPATT diode, the electric field E and
current density J are assumed to be given by:

E = E0 + E1e jω t
(2.96)
J = J0 + J1e jω t

where E0 and J0 are the DC components and E1 and J1 are the amplitudes of the AC
components (assumed to be much smaller than the DC components).
64 Microwave Devices, Circuits and Subsystems for Communications Engineering

Ca
Rd Ld Cd Rs

La
1
4
4
2
4
4
3

1
4
4
4
4
4
4
2
4
4
4
4
4
4
3
Avalanche region Drift region

Figure 2.48 IMPATT diode equivalent circuit model

Solving the device Equations (2.90) to (2.95), it is possible to obtain the linear model of
Figure 2.48.
It is composed of two sub-circuits that account, respectively, for the avalanche and drift
zones [3], and a separate loss resistance Rs. The equivalent circuit for the avalanche region
consists of a resonant circuit, with an avalanche inductance La and a capacitance Ca. The
capacitance is given by:

ε sc S
Ca = (2.97)
wa

where wa is the width of the avalanche region.


It can be shown that the device exhibits a negative resistance [1–3] for frequencies above
the avalanche resonant frequency, fa = 1/√(LaCa).
The equivalent circuit for the drift region is a series resonant circuit, with capacitance Cd
given by:

ε sc S
Cd = (2.98)
w − wa

where w is total device width.


For f > fa the resistance Rd is negative. At frequencies above fa the avalanche sub-circuit
has a capacitive behaviour. The drift region equivalent circuit, together with the avalanche
capacitance, account for the phase shift.

Self-assessment Problems
2.20 What are the two mechanisms involved in the observation of negative-resistance
in IMPATT diodes?

2.21 What is the reason for the T/4 delay between the avalanche current and the
applied sinusoidal voltage?

2.22 Why are IMPATT diodes noisy?


RF Devices: Characteristics and Modelling 65

2.9 Transistors
For the microwave circuit designer transistors are the key active elements most often used to
achieve signal generation (oscillators), signal amplification and a wide range of other, more
complex, switching and signal processing functions. (Photonic technology may come to
challenge the supremacy of transistors in the future but this is not the case yet.) Traditional
microwave transistors, including MESFETs and HEMTs, are normally constructed from
III-V group compounds. HBTs are based on both III–V and SiGe compounds.
In the remainder of this chapter we will address basic transistor modelling and review the
popular equivalent circuits used in microwave design. The principal focus will be on the
different models used in different frequency ranges and excitation (i.e. large or small signal)
amplitudes.

2.9.1 Some Preliminary Comments on Transistor Modelling


A model is a structure (physical, mathematical, circuit, etc.) that permits the behaviour of
a device to be simulated, usually under a restricted set of conditions. Such models allow
both circuit and device engineers to improve the performance of their designs.
In the context of modern microwave applications, the importance of semiconductor
modelling lies in the fact that MMIC (monolithic microwave integrated circuit) technology
does not offer the possibility of tuning once the fabrication has been completed. In this
sense, modelling is a requirement for, rather than an aid to, design.

2.9.1.1 Model types


Models can be classified in a variety of different ways. Here we will classify them according
to how they are obtained, i.e., by their extraction process.
Empirical models are obtained by describing (using mathematical functions) the macro-
scopic characteristics of devices, such as terminal currents and voltages, without regard to
the physical processes that result in these characteristics. In contrast, physical models are
derived from the known laws governing the microscopic physical process on which a device’s
behaviour ultimately depends.
Empirical models can offer the designer a level of accuracy in device behaviour prediction
approaching that of measurements. In order to obtain an empirical model, a detailed charac-
terisation of the device is required involving, for example, DC measurements, small signal
S-parameters, pulsed current/voltage measurements, etc.

2.9.1.2 Small and large signal behaviour


One method of classifying models is on the basis of their ability to correctly predict device
behaviour when the device is subjected to large input signals. Models restricted in application
to small input signals are called small signal models while those which can be applied when
large input signals are present are called large signal models.
2.9.1.2.1 Small signal models
Small signal models are widely used in microwave applications and often provide a simple,
first order, method of assessing whether a given transistor might be suitable for a particular
application of interest.
66 Microwave Devices, Circuits and Subsystems for Communications Engineering

A small signal model is valid only when the applied signal voltages consist of small fluctua-
tions about the quiescent point. It consists of a circuit that would have (to some required
level of approximation) the same S-parameters, at least over the measured frequency range,
as the device being modelled. Depending on the extraction process, the model might
not only predict the small signal behaviour of the device over the measurement frequency
band but also, to a greater or lesser extent, above and below the ends of this band. A good
model, capable of such extrapolation, is very useful when the equipment available to make
measurements does not cover the entire frequency range of interest. Individual components
in this ‘equivalent circuit’ model can often be identified with different physical processes
taking place in the device.
2.9.1.2.2 Large signal models
Large signal models are used when the assumption of small signals is invalid. They are
analytical and are able to describe the large signal properties of the device. Like the small
signal models, they consist of an equivalent circuit including non-linear components, e.g.
non-linear current sources, voltage-controlled capacitances, etc. Non-linear components,
along with appropriate characterisation procedures, allow the model user to simulate device
behaviour under different excitation conditions (transient large signal conditions, harmonic
balance conditions, etc). Small signal simulations can also be performed using this kind
of model, because linearisation of a large signal model provides the same results as a small
signal model.
The principal differences between large signal models lie in the mathematical laws chosen
to represent the non-linearities. We therefore find models due to Curtice, Materka [6] etc. for
MESFETs and HEMTs, or Gummel-Poon [8] for HBT devices, depending on the approach
that each author suggests to match the behaviour of a measured non-linearity (e.g. Ids(vgs, vds)
non-linear current source for MESFETs).
The distinguishing feature of a large signal model is its ability to correctly predict the
magnitude and shape of the signal at the device output, irrespective (within certain limits) of
the input signal’s amplitude or frequency content. The model will therefore be valid even
when the input signal exhibits large excursions about the bias, or quiescent, point.

Self-assessment Problems
2.23 What type of model would be appropriate for the design of a low noise amplifier?

2.24 What type of model would be appropriate for the design of a high power amplifier?

2.9.2 GaAs MESFETs


The GaAs MESFET (metal semiconductor field effect transistor) is widely used at micro-
wave frequencies in the realisation of oscillators, amplifiers, mixers, etc. A simplified
cross-section of a GaAs MESFET, is shown in Figure 2.49.
Three contacts can be seen on the surface of the MESFET in Figure 2.49. These contacts
are called the source, gate and drain. The source and drain contacts are ohmic while the gate
contact constitutes a Schottky diode junction. Figure 2.49 is drawn for the usual bias
RF Devices: Characteristics and Modelling 67

Vds > 0
Vgs < 0
Source Gate Drain

++++ ++++ ++
++++ ++++ ++
+++++ + n+ Contact
n+ Contact
Region Region
Depletion Region
n-Type Active Channel

Semi-Insulating GaAs Substrate

Figure 2.49 GaAs MESFET cross-section

configuration, i.e. negative gate to source voltage and positive voltage drop between drain
and source terminals.
Under normal operating conditions, the drain terminal is positively biased with respect
to the source terminal (which is often grounded), while the gate is negatively biased with
respect to both drain and source. (This implies that no significant power is sunk at the gate
terminal.) The volt drop along the conducting channel due to a current flowing from drain
to source results in the gate reverse bias being progressively greater moving from source to
drain. This creates a depletion region beneath the gate that is thicker at the drain end of the
device than at the source end.
MESFETs can also be implemented in Si technology, the principal difference between
GaAs and Si devices being operational frequency range. GaAs MESFETs can operate at higher
frequency as a result of higher carrier mobility. This makes them suitable in applications
where high operating frequency degrades the different merit figures of the transistor (e.g. gain
and noise figure). Typically, 1 GHz (often taken to mark the lower edge of the microwave
band) is the frequency above which GaAs transistors might be used with advantage.
To work properly at microwave frequencies, Gate length, Lg (see Figure 2.50), must be
less than about 1 µm in order to avoid transversal propagation effects [6, 7]. These effects
must be taken into account when wavelength becomes comparable to the physical length of
the transistor contacts. High frequency operation therefore implies small device dimensions.
Figure 2.50 shows the most important MESFET dimensions. For microwave operation,
the critical dimension is the gate length, Lg, since this determines the maximum operating
frequency. Typically, for microwave operation, this gate length is in the range 0.1–1 µm.
Another important dimension in GaAs MESFET devices is gate width. For low noise
applications, small gate widths must be employed. Large gate widths are usually reserved
for high power applications.
68 Microwave Devices, Circuits and Subsystems for Communications Engineering

Gate Length
Ls Lg Ld
a
th
id
W

Z
e
at
G

n-Type Channel

Semi-Insulating GaAs Substrate

Figure 2.50 Longitudinal view of a GaAs MESFET

Self-assessment Problem
2.25 Small gate widths are necessary in devices intended for low noise applications.
Why do you think this so?

2.9.2.1 Current-voltage characteristics


We can distinguish between two different kinds of microwave MESFET. In depletion mode
devices (which are usually used in microwave applications) any negative gate-source voltage
(greater than the pinch-off voltage) or zero gate-source voltage causes current flow from the
drain to the source. In enhancement mode devices a positive volt drop from gate to source is
necessary to allow current flow from the drain to the source. Depletion MESFETs are known
as ‘normally-off’ devices while enhancement MESFETs are known as ‘normally-on’ devices.
The voltage applied to the gate terminal acts as a control of the current flowing from
source to drain. In depletion devices, when a sufficiently negative voltage (measured with
respect to source) is applied to the gate, the depletion region extends across the entire
channel reducing drain-source current to zero. The voltage for which this just occurs is
called the pinch-off voltage. When the gate is tied to the source (i.e. gate and source voltage
are equal), the depletion region does not extend across the channel allowing drain-source
current to flow, hence its alternative description as a normally-on device.
In enhancement devices, when the gate is tied to the source, the depletion region extends
across the channel causing pinch-off of the drain-source current. In this case a positive gate
voltage (measured with respect to source) is required to shrink the depletion layer sufficiently
to open the channel for drain-source current conduction. These are called normally-off
devices.
RF Devices: Characteristics and Modelling 69

Figure 2.51 Typical I/V curves for a 10 × 140 µm GaAs MESFET device

Figure 2.51 shows a typical plot of drain-source current, Ids, versus drain-source voltage,
Vds, for several different values of gate-source voltage, Vgs. The value of Ids for Vgs = 0
is known as the saturated current level. Three characteristic operating regions for GaAs
MESFETs are shown in Figure 2.51. The saturation region starts at the Vds value where
the slope of the Ids curve (for a constant Vgs value) is zero, while the pinch-off region
refers to the area below the Vgs voltage where no significant Ids current flows for any value
of Vds.
It is sometimes desirable to know the behaviour of the derivatives of Ids with respect to Vgs
and Vds, for example, when the minimisation of intermodulation distortion in a device is
important. The output conductance of the MESFET is defined as the derivative of Ids with
respect to Vds, when Vgs is kept constant, i.e.:

 ∂I 
gds =  ds  (2.99)
 ∂vds  Vgs =constant

The inverse of Equation (2.99) is the MESFET output resistance.


The transconductance of the MESFET is defined as:

 ∂I 
gm =  ds  (2.100)
 ∂vgs  v ds =constant

The value of transconductance, and its dependence on frequency relate closely to gain that
can be realised from a circuit incorporating the MESFET and the dependence of this gain
on the frequency. Both Ids and transconductance are greatly influenced by the device’s
geometrical dimensions. The larger the gate length and gate width, the larger both drain-
source current and transconductance become.
70 Microwave Devices, Circuits and Subsystems for Communications Engineering

Self-assessment Problem
2.26 What do you think is the relationship between device geometry (i.e. physical
dimensions) and the available levels of drain current and transconductance for
high power devices?

2.9.2.2 Capacitance-voltage characteristics


On changing the bias voltage applied to the gate terminal, the charge (Qg) in the channel
beneath this terminal suffers a redistribution. As a consequence, a capacitance appears
between gate and source terminals (Cgs) and also between gate and drain terminals (Cgd). The
value of these capacitances depends on the gate-source and gate-drain voltages respectively,
thus making them (by definition) non-linear. The gate-source capacitance is defined by:

 ∂Q 
Cgs =  g  (2.101)
 ∂Vgs  vgd =constant

and the gate-drain capacitance is defined by:

 ∂Q 
Cgd =  g  (2.102)
 ∂Vgd  vgs =constant

The most important MESFET capacitance is Cgs. As would be expected, the value of Cgs is
proportional to gate area. Other things being equal, larger gate width devices therefore have
higher gate-source capacitance than the smaller gate width devices. If Cgs is high, the MESFET
gate input impedance will become low as frequency increases, and at microwave frequencies
the gate impedance may effectively become a short-circuit. Since gate terminal voltage acts
as a device gain control, a MESFET operating in this regime will no longer be useful as an
amplifying device. Small values of the Cgs are therefore necessary for MESFET operation as
a microwave amplifier.
The term unilateral, in the context of transistors, relates to the existence of output (drain)
to input (gate) feedback. A first estimate of how unilateral a device is can be deduced from
the magnitude of the S12 scattering parameter (see Chapter 3). The smaller the magnitude of
S12, the more unilateral the device.
The value of Cgd indicates whether or not a device may be unilateral. High values mean
that a device will not be unilateral while low values indicate that the device may or may not
be unilateral depending on other device parameters. Unilateral devices are required for high
frequency applications.
As for Cgs, the value of Cgd for a given gate length is directly proportional to the gate
width (Golio, 1991).
RF Devices: Characteristics and Modelling 71

Self-assessment Problem
2.27 One way of implementing a diode consists of shortening the drain and source
terminal of a MESFET. Can you imagine a way to implement a voltage controlled
capacitance?

2.9.2.3 Small signal equivalent circuit


The most widely used circuit model for the GaAs MESFET is shown in Figure 2.52.
The parasitic inductances, Lg, Ld and Ls, originate in the metallic contacts deposited on the
device surface and any associated bonding wires. The value of these elements depends on
the layout of the device, as well as the properties of its constituent materials. In most cases
Lg is the largest inductance and Ls the smallest.
The origin of the parasitic resistances Rs and Rd, is the ohmic contact of drain and source
terminals as well as a small contribution of the resistance due to the active channel. The
resistance Rg represents the metalisation resistance resulting from the Schottky gate contact.
The capacitances Cpgi and Cpgi arise from device packaging and may be important when the
operating frequency of the MESFET is high. The capacitance Cds accounts for coupling
between drain and source terminals and is assumed to be bias independent. The equivalent
circuit contains five non-linear components. These are:
Cgs used to model gate charge dependence on Vgs.
Cgd used to model gate charge dependence on Vgd.
Igs used to model the gate-source diode.
Igd used to model the gate-drain diode.
Ids governed by Vgs and Vds and used to model the primary non-linear element from which
the amplification properties of the device are derived.

Igd

Igs
Ids
(= gmVgs )

Figure 2.52 Equivalent circuit model for GaAs MESFET


72 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 2.53 Small signal equivalent circuit for GaAs MESFET

For small signals the equivalent circuit of Figure 2.52 can be transformed, for a given bias
point, to the linearised model shown in Figure 2.53. Here, Cgs and Cgd have been fixed at
their value appropriate to the chosen bias point, and the non-linear gate-source and gate-
drain diodes have been replaced with linear resistors rgs and rgd. The values of these resistors
are, of course, those appropriate to the bias conditions, i.e.:

1
rgd = (2.103)
 ∂Igd 
 ∂v 
 gd 

1
rgs = (2.104)
 ∂Igs 
 ∂v 
 gs 

The small signal transconductance, gm, and output conductance, gds, are given by:

 ∂I 
gm =  ds  g(2.105)
 ∂vgs  Vds =constant

 ∂I 
gds =  ds  g(2.106)
 ∂vds  Vgs =constant

The value of gm is related to the small signal gain at any given bias point. Knowledge of
the dependence of gm on frequency allows the prediction of small signal device gain as a
function of operating frequency. Furthermore, the change in gm with Vgs and Vds (i.e. changes
in bias point) can be identified.
RF Devices: Characteristics and Modelling 73

gds depends (similarly to gm) on both frequency and bias point. It is related to the DC
output resistance, r0, of the device by:

1
r0 = (2.107)
gds

Another important device parameter is transconductance time delay, sometimes referred to


more simply as time delay. Due to the time it takes for charge to redistribute itself when
a voltage gate change occurs, the transconductance is not able to instantaneously follow
fast voltage changes. A delay time is therefore added to the transconductance behaviour in
order to obtain a more complete representation of the ‘gain process’. This delay has to
be taken into account at high operating frequencies. A typical value of this delay time for
GaAs MESFET devices is between 1 ps and 3 ps.
For microwave operation, the equivalent circuit shown in Figure 2.53 can be obtained
from S-parameter measurements, along with linear model extraction techniques developed
by several authors [6, 7]. The complexity of these extraction methods is due to the difficulty
of separating effects and identifying the value of the different elements when the operating
frequency becomes high. As an additional handicap, gm and gds have both been found to be
frequency dependent [6]. To give an idea of the complexity of the extraction process, con-
sider how the decreasing impedance of Cds can obscure the variation of the transconductance
with frequency.
Some second-order effects, e.g. trapping [6, 7, 8] cause a frequency dispersion to appear
around tens of kilohertz. This dispersion can be observed, from a macroscopic point of
view, as a sudden change in both gm and gds from their DC values to their RF values. Several
authors propose ways of modelling these effects. These second-order phenomena will not
be treated here. It is important to be aware, however, that to correctly design circuits for
operation at, or close to, the frequency where the dispersion appears these effects must
generally be considered. When designing small signal broadband amplifiers, for example,
the amplifier gain is proportional to gm. If the frequency band of the amplifier to be designed
covers the region of tens of kilohertz, care must be taken in the design process because the
transconductance may exhibit variation at these frequencies, the gain of the amplifier being
modified as a consequence.
In general, all the element values in the equivalent circuit model of a MESFET display
a dependence on the geometric dimensions of the device. The relationships between device
size and the value of the model elements are known as scaling rules. These rules are useful
in estimating the value of the element values of the small signal model when direct measure-
ment of them is difficult or impossible.

Self-assessment Problem
2.28 Calculate the small signal low frequency voltage gain of a GaAs MESFET
(using the circuit in Figure 2.53) by ignoring all the parasitic resistors, inductor
and capacitors.
74 Microwave Devices, Circuits and Subsystems for Communications Engineering

2.9.2.4 Large signal equivalent circuit


The circuit shown in Figure 2.52 can be used to model the behaviour of a GaAS MESFET
under any practical operating condition. It is therefore valid for DC, RF small signal and RF
large signal conditions.
A large signal model consists of the values of the linear elements in the equivalent
circuit along with the equations that define the non-linear elements. For the model shown
in Figure 2.52 the latter comprise the current of the non-linear source, Ids, as a function of
vgs and vds and the capacitance of the non-linear elements Cgs and Cgd as a function of their
charge control voltages. (The equations defining the non-linear gate-source and gate-drain
current sources are implicit in their representation as diodes.)
A variety of models can be found in the literature [6]. The optimum model choice depends
on several different criteria such as accuracy, complexity and CPU time required to run the
model etc. One of the most widely applied GaAs MESFET models, however, is that due to
Curtice [6]. This incorporates an analytical expression for Ids that closely approaches the
measured behaviour of this non-linear element.
The general process of extracting a non-linear large signal model consists of fitting the
appropriate measurements to a set of expressions, by optimising the expression parameters.
Measuring the gate-source and gate-drain junction currents for a set of forward bias
conditions it is possible to characterise the non-linear current sources that characterise these
junctions. The forward currents Igs and Igd are usually modelled by the expressions:

Igs = Inss (eαVgs − 1) (2.108)

Igd = Insd (eαVgd − 1) (2.109)

In order to model reverse current breakdown, a diode-like current source is used. The break-
down current, Id,br, (present when vd < −Vbr) is given by:

 Is (e −( vgs −Vbr )α ) for vgd < −Vbr


Igd ,br =  (2.110)
 0 for vgd > −Vbr

where Vbr is the breakdown voltage.

2.9.2.5 Curtice model


The GaAs MESFET model that is currently most widely known was proposed by Curtice [6]
and focuses on the non-linear elements Cgs, Cgd and Ids.
In the expressions that follow the independent variables are intrinsic (or internal) voltages
Vgi and Vdi. To find the dependence of the non-linear elements on Vgi and Vdi the value of the
source, drain and gate resistances (Rs, Rd and Rg) must be known. With these resistance
values it is possible to evaluate the intrinsic voltages using:

Vgs = Ig Rg + Vgi + (Ig + Id)Rs (2.111)

which gives Vgi, and:

Vds = Id Rd + Vdi + (Ig + Id)Rs (2.112)


RF Devices: Characteristics and Modelling 75

which gives Vdi. (Here we have assumed there is no breakdown effect and only forward
gate-source and drain-source currents are present.)
The dependence of Ids on the intrinsic voltages Vgi and Vdi is given by:

Ids(Vgi, Vdi) = β(Vgi − VTO)2(1 + λVdi) tanh (αVdi) (2.113)

where α, β, λ and VTO are the model parameters. The modelling process for Ids consists
of finding the values for the parameters of Equation (2.113) such that the equation best
represents the real behaviour (i.e. the measured I/V characteristic) of the device.
The dependence of capacitances Cgs and Cgd on vgi and vdi in the Curtice model, obtained
from the classical theory of Schottky junctions, is given by:
−1 / 2
 V 
Cgs = Cgso 1 − gi  (2.114)
 Vbi 
−1 / 2
 V 
Cgd = Cgdo 1 − di  (2.115)
 Vbi 

where Cgso, Cgdo are the non-linear capacitances at zero bias voltage, Vbi is the built-in voltage,
and Vgi, Vdi are the intrinsic voltages.
From Equation (2.110) it is possible to obtain analytical expressions for gm and gds as a
function of the intrinsic voltages. Using Equations (2.105) and (2.106) these expressions are:

∂Ids
gm = = Ids[2/(Vgi − VTO)] (2.116)
∂Vgi

∂Ids
gds = = β (Vgi − VTO)2(1 + λVdi){α /[cosh2(αVdi)]} +
∂Vi
+ β (Vgi − VTO)2λ tanh (αVdi) (2.117)

Many more models for GaAs MESFET devices are described in the literature.

Self-assessment Problem
2.29 How can a MESFET large signal model provide information about harmonic
content when injecting a signal of frequency fo through the gate terminal? (Use
the Curtice model and consider the principal non-linearity to be drain current.)

2.9.3 HEMTs
The high electron mobility transistor (HEMT) is a type of field effect transistor. Unlike
GaAs MESFETs, however, HEMT devices are heterostructures and as such are able to
operate at higher frequencies. Figure 2.54 shows a cross-section of a typical HEMT device.
It has three metal contacts at the gate, drain and source terminals. The source and drain
76 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 2.54 Cross-section of a HEMT device

terminals are ohmic contacts. The gate contact constitutes a Schottky junction. In these
respects it is identical to a GaAs MESFET.
The improved performance of HEMTs over MESFETs comes about due to the higher
value of the electron mobility (and therefore velocity) in these devices. This makes HEMTs
suitable for applications where low noise and/or high frequency of operation are required.
Several different HEMT physical structures are possible [6, 7]. Each topology, or structure,
has advantages in terms of some specific HEMT feature such as power dissipation capability,
maximum operating frequency, noise performance, etc. and many commercial foundries offer
an advice service to circuit designers on which HEMT topology will be most appropriate to
a give device application.
The most important geometric dimension in an HEMT device (as for the GaAs MESFET)
is gate length, which limits the maximum operating frequency of the device. Typical gate
lengths for HEMTs are similar to those found in MESFETs. A further feature determining the
general behaviour of the HEMT, however, is the thickness of the undoped, and the n-type,
AlGaAs layers (see Figure 2.54). Typically, the thickness of the n-type AlGaAs layer may be
between 0.03 and 0.2 µm. A typical thickness for the undoped layer (also known as spacer)
might be 0.005 µm.

2.9.3.1 Current-voltage characteristics


From a macroscopic point of view, I/V characteristics of MESFETs and HEMTs are quite
similar. The drain-source current (Ids) dependence on gate width observed in MESFETs can
also be observed in HEMTs.
A particular HEMT feature can be observed in the I/V output characteristics illustrated in
Figure 2.55. As gate-source voltage (for constant vds) increases beyond a certain point, the
corresponding increases in Ids become smaller. This represents a degradation, or compression,
of the device’s transconductance, gm (see Figure 2.56). Transconductance compression is an
HEMT feature not found in MESFETs.
RF Devices: Characteristics and Modelling 77

Figure 2.55 Typical I/V curves of a 6 × 150 µm HEMT device

Figure 2.56 Typical gm versus vgs curves for a 6 × 150 µm HEMT

The same definitions of transconductance and output resistance as used for MESFETs can
be applied to HEMTs. Due to its physical structure, however, the output conductances of
HEMTs are higher than those observed for MESFET devices. Similar to MESFET devices,
HEMT output resistance is inversely proportional to device gate width and directly propor-
tional to gate length. Finally, transconductance is proportional to gate width, and inversely
proportional to gate length.
78 Microwave Devices, Circuits and Subsystems for Communications Engineering

Self-assessment Problem
2.30 Suppose you are using a HEMT device to implement a small signal LNA and,
due to the low level of signal to be received you have to provide high gain.
Looking at Figures 2.55 and 2.56, comment on the region you would choose in
which to bias the device.

2.9.3.2 Capacitance-voltage characteristics

The following two non-linear capacitances, Cgs and Cgd, are important in HEMT devices:

 ∂Q 
Cgs =  g  (2.118)
 ∂Vgs  vgd = constant

 ∂Q 
Cgd =  g  (2.119)
 ∂Vgd  vgs = constant

(These definitions are the same as those applied to MESFET devices.) A study of the
dependence of Cgs and Cgd on HEMT dimensions suggests that Cgs is proportional to gate
area, while Cgd is proportional to gate width.

2.9.3.3 Small signal equivalent circuit

The same equivalent circuit model presented for MESFETs can be used for the HEMT (see
Figure 2.53). The significance and interpretation of the equivalent circuit elements are the
same as for MESFET devices although the element values, and the expressions used to
model the non-linear elements, naturally differ.

2.9.3.4 Large signal equivalent circuit

As in the case of the small signal model, the circuit topology for the HEMT large signal model
is identical to the corresponding equivalent circuit of a GaAs MESFET (see Figure 2.52).
The element values and the expressions used to model the non-linear current sources and
capacitances are, naturally, different, however.
As for MESFETs, the process of extracting a non-linear large signal model consists in
fitting a set of expressions to a set of measurements by optimising the expressions’ coefficients.
The forward diode currents are well modelled by:

Igs = Inss (eαVgs − 1) (2.120)

Igd = Insd (eαVgd − 1) (2.121)


RF Devices: Characteristics and Modelling 79

and the breakdown current is given by:

 Is (e −( vgs −vbr ) ) for vgd < −Vbr


Igd , br =  (2.122)
 0 for vgd > −Vbr

where Vbr is the breakdown voltage.


HEMT transconductance begins to degrade [6, 7], when a particular value of vgs, often
denoted by vpf , is reached. The large signal model must take this degradation into account to
accurately predict the behaviour of the device. To accurately model this effect, many models
have been proposed [6]. As in the case of the MESFET, here we present one model as an
example. To establish a point of comparison with the MESFET the example chosen is the
HEMT Curtice model proposed by [6].
It is possible to represent the non-linear current source, Ids, for HEMT devices in terms of
the MESFET current source model, IdsFET, i.e.:

 IdsFET for Vgi ≤ Vpf



Ids =  ξ (2.123)
 IdsFET − ψ + 1 (Vgi − Vpf ) tan(αVdi )(1 + λVdi ) for Vgi > Vpf
ψ +1

 0 for Vgi ≤ Vp0


IdsFET =  (2.124)
β (Vgi − Vp0 )2 tanh(αVdi )(1 + λVdi ) for Vgi > Vp0

and α, β, λ, ξ, ψ, Vpf and Vto are the non-linear model parameters, Vpf being the gate-source
voltage for which the transconductance degradation appears.
The intrinsic voltages Vgi and Vdi (see Figure 2.52) may be calculated using:

Vgi = Vgs − Ig Rg − (Ig + Id)Rs (2.125)

Vdi = Vds − Id Rd − (Ig + Id)Rs (2.126)

The non-linear capacitances, Cgs and Cgs, are quite different from those observed in MESFET
devices. [6], for example, adapts capacitance expressions taken from MEYER’s empirical
MOSFET model [10] resulting in the following expressions:

2  (Vdss − Vdi )2 
 1 − CG + cGS 0 for Vdi < Vdss
3 (2Vdss − Vdi )2 
Cgs =   (2.127)
 2
 CG + CGS 0 for Vdi ≥ Vdss
3

2  2
Vdss 
 1 − C + cGS 0
2 G
for Vdi < Vdss
Cgd =  3  (2Vdss − Vdi )  (2.128)

 CGS 0 for Vdi ≥ Vdss
80 Microwave Devices, Circuits and Subsystems for Communications Engineering

where:

Cm 0 (Vgs − VT 0 )
1/ X
for Vgi > VT 0
CG =  (2.129)
 0 for Vgi ≤ VT 0

and:

 V 
Vds 0 1 − gi  for Vgi > VT 0
Vdss =  V TO  (2.130)
0 for Vgi ≤ VT 0

Other approaches to modelling both the drain-source non-linear current source and the non-
linear capacitances can be found in the literature, e.g. [6, 7]. The choice of a particular model
depends on the application (and often, therefore, the circuit topology), the trade-off between
model complexity and accuracy, and the effort required to obtain the model parameters.

Self-assessment Problem
2.31 Do you think the parasitic resistor Rg will affect the use of both MESFETs and
HEMTs as low noise amplifiers? Explain your answer.

2.9.4 HBTs
The heterojunction bipolar transistor (HBT) has high transconductance and output resistance,
high power handling capability and high breakdown voltage. The use of bipolar transistors
in microwave applications is not common due to the limitation on their transition frequency.
It can be shown that the transition frequency of a bipolar device depends on the base resist-
ance; the higher the base resistance, the lower the transition frequency. A critical advantage
of HBTs compared with BJTs is that they have very high base doping and hole injection into
the emitter is suppressed. In this way it is possible to obtain low base resistance combined with
wide emitter terminal dimensions and, as a consequence, very high operating frequency can
be achieved. This makes these devices attractive for high power microwave and millimetre-
wave applications. The principal advantages of HBTs over traditional BJTs, GaAs MESFETs
and HEMTs can be summarised as follows:

1. Base resistance can be reduced as a consequence of higher base doping.


2. Base-emitter capacitance is reduced as a result of lower emitter doping.
3. Low transit time values due to the use of AlGaAs/GaAs material.
4. Low output conductance value.
5. High transconductance values.
6. High gain.
7. High breakdown voltages.
8. Low 1/f noise due to the absence of second-order effects appearing in GaAs MESFETs
and HEMTs.
RF Devices: Characteristics and Modelling 81

b b
+ ++
c
+ ++ n-
+ ++ + + +
+ ++ + + +
+ ++ + + +
+ ++
SiGaAs + + +

Impurity

Figure 2.57 Cross-section of a AlGaAs/GaAs device

9. High transition frequencies due to the value of base resistance achieved.


10. High current/power handling capability.

The dominant factors in the rapid progress of HBT devices are improvements in the quality
of materials and improvements in epitaxial layer growth techniques (e.g. molecular beam
epitaxy and metallic organic chemical vapour deposition). Recently a new technology based
on SiGe material has allowed HBT devices to be realised with the same RF and DC properties
as III–V group based devices, decreasing the manufacturing cost as well as the technological
process complexity.
We now outline a generic HBT model that, at least for first-order analysis, can be applied
to both traditional III–V HBTs or to SiGe HBTs, only the element values varying for the
different HBT materials.
A cross-section of an AlGaAs/GaAs HBT is shown in Figure 2.57. (In this particular
device an InGaAs cap is used to obtain low contact resistance values. Berilium, Be, has been
used as material for the base terminal. Base thickness can be reduced using carbon-doping
techniques, this being a good way to reduce base resistance, thus achieving higher transition
frequency.

Self-assessment Problem
2.32 From a manufacturing point of view do you think SiGe HBT devices present any
advantages over MESFETs and HEMTs? (Hint: Think in terms of the integration
of analogue and digital systems.)

For the HBT, in contrast to MESFETs and HEMTs, there are not several different equivalent
circuit models to predict behaviour under different operating conditions (small signal, large
signal, etc.). We will therefore discuss the basic operation of the HBT device, presenting at
the same time the Gummel-Poon [8] model that may be extended to represent the main
non-linear elements. This model is not, in itself, able to predict some second-order effects
such as I/V-curve dependence on device self-heating for high values of base-current bias
(important for high-power HBTs) and collector junction breakdown. The Gummel-Poon
model is both the original and an interesting approach to HBT performance prediction and is
82 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 2.58 Large signal equivalent circuit for HBT

also useful as a starting point for understanding the device’s physical behaviour. Other HBT
models that attempt to account for effects not predicted by the Gummel-Poon model are
discussed in [7].
A complete non-linear equivalent circuit model of an HBT is shown in Figure 2.58.
Taking this model as a starting point, it is possible to explain the different operating condi-
tions that can be observed in an HBT device. The study that we are going to carry out, under
DC operation, is the same as proposed for BJT devices by other authors [8]. This approach
can be justified taking since the important differences between BJT and HBT devices become
apparent only at high frequencies (where the superior performance of the HBT over BJT
allows its use in microwave applications), the DC behaviour being nearly the same for both
devices.
Under forward bias conditions (Vbe ≥ 0 and Vbc < 0) device behaviour is well predicted by
the simplified equivalent circuit shown in Figure 2.59.

Figure 2.59 DC HBT equivalent circuit under forward bias


RF Devices: Characteristics and Modelling 83

Figure 2.60 DC HBT equivalent circuit under reverse bias

The forward bias DC current sources proposed by Gummel-Poon [7, 8] are given, respect-
ively by:

  V  
Icc = Isf  exp  be  − 1 (2.131)
  n f KT  
and:

  V  
Ibe = Ise  exp  be  − 1 (2.132)
  ne KT  

The Icc current source represents the collector current while base current is represented by
two sources, one of them depending on βf, the forward DC gain of the device.
Under reverse bias conditions device behaviour is predicted by the equivalent circuit
shown in Figure 2.60.
The reverse bias DC current sources proposed by Gummel-Poon [8] are, respectively,
given by:

  V  
Iec = Isr  exp  bc  − 1 (2.133)
  nr KT  

  V  
Ibc = Isc  exp  bc  − 1 (2.134)
  nc KT  

As for MESFETs and HEMTs, the resistances Rb, Rc and Re play an important role in the
HBT modelling process. This is because when we measure the I/V curves of an HBT device
we represent the measured collector current versus the external, or applied, voltages Vc and
Vb. However, the voltages appearing in Equations (123)–(126) are intrinsic, or internal,
voltages. Knowing Rb, Rc and Re the intrinsic voltages can be calculated as follows:

Vbe = Vb − Ie Re − Ib Rb (2.135)

Vbc = Vb − Vb + Ic Rc − Ib Rb (2.136)

These resistances can be estimated from geometric and material considerations [7, 8] or
from device measurements [8]. (The former method requires knowledge of some fabrication
84 Microwave Devices, Circuits and Subsystems for Communications Engineering

process parameters while the latter requires a complex measurement set-up, not available in
most microwave laboratories, to carry out the device characterisation.)

Self-assessment Problem
2.33 What type of transistor would you select for a communication system in which
minimisation of power consumption was a prime consideration? Explain your
answer.

2.9.4.1 Current-voltage characteristics


Figure 2.61 shows a typical (grounded emitter) HBT I/V curve.
The currents in Figure 2.61 are governed by Equations (2.131)–(2.134).
More complex expressions for the HBT currents, taking into account several second-order
effects such as self-heating, breakdown, etc., can be found in the literature, e.g. [7]. In most
cases, however, the Gummel-Poon model is adequate. (The accuracy of this model will
depend, of course, on how the different parameters have been extracted.)

2.9.4.2 Capacitance-voltage characteristics


Two different kinds of capacitances can be distinguished in HBT devices; depletion
capacitances and diffusion capacitances.

Figure 2.61 Typical I/V curve of a HBT device


Note: DEVICE: H67 HBT from Daimler-Benz Ib = 0.1 mA to 1.1 mA, increment 0.1 mA
RF Devices: Characteristics and Modelling 85

2.9.4.2.1 Depletion capacitance


In HBT devices there are two intrinsic depletion, or junction, capacitances; the base-
collector capacitance, Cbc, and the base-emitter capacitance, Cbe. As in the case of MESFETs,
the origin of these capacitances is in the electron and hole concentration changes as a result
of base-emitter and/or base-collector voltage changes. Cbe, which lowers the maximum operat-
ing frequency of the HBT, can be reduced by decreasing either the collector thickness or the
base-collector area. The junction capacitances are given by the following expressions:

Cbeo
Cbe = (2.137)
V
1 − be
Vbuilt

Cbeo
Cbe = (2.138)
V
1 − be
Vbuilt

where Cbeo and Cbco are the intrinsic capacitances when the applied voltages are zero, and
Vbuilt is the barrier height voltage.

Self-assessment Problem
2.34 Explain briefly why the capacitance Cbe may limit HBT device performance
depending on the operating frequency.

2.9.4.2.2 Diffusion capacitances


Electrons present in the base are due to diffusion current. The total base electron concentra-
tion near the collector defines the collector current. A direct relation between changes in
electron concentration and changes in collector current therefore exists which means that a
diffusion capacitance can be defined at the base terminal. For large forward bias, diffusion
capacitance (in addition to depletion capacitance) must be considered.
A base diffusion time, τb, can be calculated using:

Wb2
τb = (2.139)
2 Dn

where Wb is base thickness and Dn is the diffusion constant for electrons in the base. For the
collector, we can define a transit time given by:

X
τc = (2.140)
2vsat

where X is the collector thickness and vsat is the electron saturate velocity. The transition
frequency, ft, of the HBT is then given [7] by:

1
ft = (2.141)
2πτ ec
86 Microwave Devices, Circuits and Subsystems for Communications Engineering

where:

τec = re + (Cbe + Cbc) + τb + τc + Cbc(Rc + Re) (2.142)

and the emitter junction resistance, re, in Equation (2.142) is:

∂Vbe n f KT
re = ≈ (2.143)
∂Ie Ie

The base-emitter diffusion capacitance can be calculated from:

τc + τb
Cbet = (2.144)
re

where:

Wb2  I* 
τb = + τ bX 1 − c  (2.145)
2 Dn  Ic 

and I*c is the value of collector current corresponding to the minimum of the τec versus 1/Ic
plot. (N.B. the (correction) term containing τbx is only applied when Ic > I*c .)
The above equations express an increase of τec for low values of 1/Ic due to the hole and
electron injection into the collector. This is known as the base widening effect and can be
modelled by replacing the base width by an effective width equal to Wb + X where X is
collector thickness. This results in an effective diffusion capacitance that can be calculated
from Equation (2.144).

2.9.4.3 Small signal equivalent circuit


Figure 2.62 shows the small signal equivalent circuit of an HBT based on a one-dimensional
transistor model [7].
Other equivalent circuit topologies that model the small signal behaviour of the transistor
are possible and some of these, extracted from geometrical parameters and manufacturing
information, are quite different from those applied to BJTs. Such models may approximate the
device performance very well including second-order effects not predicted by the equivalent
circuit of Figure 2.62.
The admittance parameters [Y] of the small signal equivalent circuit of Figure 2.62
can be calculated analytically and, once known, these can be transformed into scattering
(s-) parameters (see Section 3.8.2, Chapter 3). The value of the different elements of the
equivalent small signal circuit can then be found using linear extraction techniques [7]. The
values obtained can be refined by applying an optimisation process to improve the model’s
accuracy (i.e. make the model’s S-parameters match those of the device which would be
measured more closely).
A wide variety of small signal HBT models can be found in the literature, e.g. [7, 8].
The appropriate choice of model depends on the application and on the modelling facilities
available by the user. The trade-off between the simplicity of the extraction process and the
simplicity with which the final model can be used may also be a consideration.
RF Devices: Characteristics and Modelling 87

Figure 2.62 Small signal HBT equivalent circuit model

2.9.4.4 Large signal equivalent circuit


The most widely used large signal circuit model for the HBT is that shown in Figure 2.58.
This model is implemented in most of the non-linear circuit simulation packages (e.g. MDS,
LIBRA, PSPICE). Other model topologies have been proposed in the literature, e.g. [7].
The procedure to extract the large signal model is the same as for MESFETs and HEMTs.
Forward and reverse bias current source expressions, found from a set of forward and
reverse bias DC measurements [7, 8] are given by Equations (2.125)–(2.128).
Since the final DC model must predict the collector current I/V curves, a parameter refine-
ment procedure involving optimisation methods [7] may be necessary. In this case a set of Ic
versus Vc curves (emitter grounded) with Ib as parameter are measured experimentally, along
with Vb. From the circuit in Figure 2.58 and Equations (2.125)–(2.128) the following current
balance expressions can be found:

  qV    1   qV     qV  
Ic = Isf exp  be  − 1 − Isr 1 +  exp  bc  − 1 − Isc exp  bc  − 1 (2.146)
  n f KT    βr    nr KT     nc KT  

Isf   qVbe     qV   I   qV  
Ib = exp   − 1 + Ise exp  be  − 1 + sr exp  bc  − 1
β f   n f KT     ne KT   βr   nc KT  
  qV  
+ Isc exp  bc  − 1 (2.147)
  nc KT  
The internal voltages Vbe and Vbc are calculated from the external Vb and Vc voltages using:

Vbe = Vb − (Ib + Ic)Re − Ib Rb (2.148)


88 Microwave Devices, Circuits and Subsystems for Communications Engineering

Vbc = Vb − Vc − Ib Rb (2.149)

It is then possible to optimise the value of the parameters in Equations (2.146) and (2.147)
to fit the measurements, thus obtaining the large signal model of the non-linear current
source, Ic.
In the case of the non-linear capacitances, the process consists of choosing the parameters
of Equations (2.137)–(2.145) to get good agreement between measurements and model pre-
dictions. In this process we take as experimental data the results obtained from the extraction
of the HBT small signal model at each different bias point, as described for MESFET’s in
Section 2.9.4.3.

2.10 Problems
2.1 What are the phenomena giving rise to the two parallel capacitors in the equivalent
circuit model of the PN diode?
2.2 Calculate the conversion losses in a varactor multiplier with input frequency 4 GHz.
The diode parameters are I0 = 10−12 A, α = 40 V−1 and γ = 0.33.
2.3 What are the phenomena giving rise to the formation of the potential barrier in the
Schottky diode?
2.4 A Schottky diode, connected in parallel across a transmission line, is used for the
detection of a microwave signal at 9 GHz. The diode characteristics are I0 = 10−10 A,
α = 40 V−1. The amplitude of the microwave signal is 7 mV. Calculate the amplitude
of the detected DC current.
2.5 What are the three main effects resulting from the presence of a long intrinsic region
embedded in a PIN diode?
2.6 Obtain the possible phase-shift values between input and output provided by the circuit
of Figure 2.36, when three PIN diodes are used with transmission line sections of
length: lx = λ g/10, l1 = 3λ g/20, l3 = λ g/4.
2.7 What should be the length of a Gunn diode operating in dipole layer mode at 10 GHz?
2.8 A switch is manufactured by shunting a PIN diode across a 50 Ω transmission line.
The diode has a length d = 100 µm, an average carrier lifetime τ = 6 µs and a mobility
µ = 600 cm2 V−1 s−1. The simplified model includes a variable resistor, along with an
intrinsic capacitor, CI = 0.1 pF, and a loss resistance, Rs = 0.5 Ω. Calculate: (a) the
insertion losses for a bias current Id = 0 mA; and (b) the isolation for a bias current
Id = 12 mA.
2.9 A Gunn diode has the simplified non-linear model given by a non-linear conductance
G(V) = −15 + 2V mS and a parallel capacitance value C = 2 pF. Design: (a) a stable
negative resistance amplifier at 9 GHz; and (b) a stable free-running oscillator with
maximum output power at 9 GHz.
2.10 Sometimes, when designing mixers for high frequency communication systems, GaAs
MESFETs are employed as resistive elements. Given that the most common mixer
structure includes diodes as non-linear elements, explain how to bias a GaAs MESFET
to be used in the linear region as a resistor. (NB Gate-source and gate-drain junctions
are Schottky junctions. In the I/V GaAs MESFET curves three, well-differentiated,
operating regions can be distinguished.)
RF Devices: Characteristics and Modelling 89

2.11 Suppose you are designing a single stage small-signal amplifier based on a GaAs
MESFET device. Using the small-signal equivalent circuit MESFET model and con-
sidering Ri = Rs = 0 Ω, Ls = 0 H, discuss qualitatively the elements of the equivalent
circuit responsible for the gain degradation of the amplifier with increasing the frequency.
What is the influence of Rs and Ls in the particular case of Rs ≠ 0 and/or Ls ≠ 0
(maintaining Ri = 0), on the behaviour of the device as an amplifier?
2.12 Consider the MESFET equivalent circuit model. The Ids parameters for the Curtice
model presented for a GaAs MESFET device (Equation (2.113)) are β = 0.12 mA,
VTO = −2.5 V, λ = 0.4 × 10−2 V−1 and α = 2.34 V−1. The Cgs and Cgd non-linear
capacitances (Equations (2.114) and (2.115)) are known with Cgso = 2.3 × 10−11 F,
Cgdo = 0.02 × 10−12 F and Vbi = 0.7 V. The rest of the elements can be assumed to
be open circuits (capacitances) or short circuits (resistances and inductances). For a
gate-source bias of vgs = −0.25 V and a drain-source bias of vds = 3 V, calculate the
small-signal equivalent circuit model of the device. (Note that neither the forward
gate-source current nor the breakdown effect need be considered.)
2.13 When designing oscillators for digital communications systems, a critical figure of
merit is the oscillator phase noise. Analogue oscillators are based on transistors as
the active devices responsible for oscillation. The phase noise can be viewed as a
low frequency noise, very near to the carrier, that degrades the device behaviour.
Which of the three transistor types discussed in this chapter would be most suitable
for application in a reference oscillator? Explain your answer. (Hint: think about the
flicker, i.e. 1/f, noise.)

References
[1] S.M. Sze, Physics of Semiconductor Devices, John Wiley & Sons, New York, 1981.
[2] A. Vapaille and R. Castagne, Dispositifs et circuits intégrés semiconducteurs, Dunod Université, Bordas,
Paris, 1990.
[3] K. Chang, Microwave Solid-State Circuits and Applications, Wiley Series in Microwave and Optical Engineer-
ing, John Wiley & Sons, New York, 1994.
[4] P.F. Combes, J. Graffeuil and J.F. Sautereau, Microwave Components, Devices and Active Circuits, John
Wiley & Sons, New York, 1988.
[5] E.A. Wolff and R. Kaul, Microwave Engineering and System Applications, John Wiley & Sons, New York,
1988.
[6] J.M. Golio, Microwave MESFETs and HEMTs, Artech House, Norwood, MA, 1991.
[7] R. Anholt, Electrical and Thermal Characterization of MESFET’S, HEMT’s and BT’s, Norwood, MA, Artech
House, 1995.
[8] R.S. Muller and T.I. Kamins, Device Electronics for Integrated Circuits, John Wiley & Sons, Inc., New York,
1982.
[9] K.V. Shalimova, Semiconductors Physics, MIR, 1975.
[10] J. Meyer, ‘MOS models and circuit simulation’, RCA Rev, vol. 32, March 1971, pp. 42–63.
90 Microwave Devices, Circuits and Subsystems for Communications Engineering
Signal Transmission, Network Methods and Impedance Matching 91

3
Signal Transmission, Network
Methods and Impedance Matching
N. J. McEwan, T. C. Edwards, D. Dernikas and I. A. Glover

3.1 Introduction
Signal transmission, network methods and impedance matching are all fundamental topics
in RF and microwave engineering. Signals must be transmitted between devices such as
mixers, amplifiers, filters and antennas, and at frequencies where wavelength is comparable
to, or shorter than, the separation of these devices, transmission line theory is required to
design the connecting conductors properly. There are several technological implementations
of transmission lines. The most familiar, in the RF context, is coaxial cable and this techno-
logy is still important (with others) where long and/or flexible lines are required. The most
important transmission line for shorter distances, where rigidity is not a disadvantage, is
microstrip and this technology is, therefore, given particular attention in this chapter.
Network methods refer to the collection of mathematical models that relate the electrical
quantities at the ports (inputs and outputs) of a device. The device may be passive (such as a
piece of transmission line) or active (such as an amplifier). Usually, but not always, devices
have two ports: an input and an output. Some two-port descriptions may be familiar from
other applications, e.g. h-parameters for lower frequency electronics and ABCD-parameters
for power systems. The two-port description used at RF and microwave frequencies employs
s-parameters and it is these on which this chapter therefore concentrates. There are good
practical reasons for adopting different parameter sets for different applications (to do with
ease of measurement and/or ease of use) but all such sets contain equivalent information about
the device and translation between sets is straightforward. The utility of network parameters
is that they characterise the systems aspects of a device or subsystem completely – but free
from the potentially distracting details of how the device or subsystem works.
Impedance matching (or impedance compensation) is important because it affects the
way devices interact when they are connected together. It consists of altering the input or
output impedance of a device to make it more compatible in some way with another device
to which it is to be connected. It may be designed to realise one of several quite different
objectives either at a single frequency or over a band of frequencies (e.g. minimising reflected

Microwave Devices, Circuits and Subsystems for Communications Engineering Edited by I. A. Glover, S. R. Pennock
and P. R. Shepherd
© 2005 John Wiley & Sons, Ltd.
92 Microwave Devices, Circuits and Subsystems for Communications Engineering

power, maximising power transfer, maximising gain or minimising noise figure). Alternatively,
matching may be designed to achieve some satisfactory compromise between more than one
of these objectives.
Loosely speaking, signal transmission is the process by which signals are transferred by
transmission lines from one device another, network methods describe the change in a signal
as it enters, traverses and leaves a device and impedance matching is the technique used to
optimise the overall desired characteristics of the device and its associated input and output
transmission lines. This collection of technologies and techniques represents a basic tool kit
for the RF and microwave design engineer.

3.2 Transmission Lines: General Considerations


A transmission line is a structure that is used to guide electromagnetic waves along its length.
The most obvious practical purpose is either to transport power from place to place, as, for
example, in power cables, such as overhead lines on pylons, or to transport information –
but to transmit information, some energy must be transported in any case.
In RF engineering, a third and extremely important function is as circuit elements, for
example, in impedance transforming networks, or in filters. This function is based on the ability
of transmission lines to store energy, as more familiar circuit elements such as inductors and
capacitors do.

3.2.1 Structural Classification


Transmission lines have an immense variety of forms, and it is important to realise that they
need not conform to elementary ideas of an electric circuit. Here we are restricting ourselves
to a specialised subset, which can be at least partially understood from a circuit point of
view. Before focusing on this, we need to set this group within the context of more general
structures, and to see the features that distinguish it.
A transmission line does not need to have a uniform cross-sectional structure along its
length. There are structures, for example, that use periodic corrugations to guide waves along
their surface. Our first stage of specialisation, therefore, is to consider only structures that
are longitudinally homogeneous. This still leaves, however, a great variety.
Consider Figure 3.1 that sets out examples of important transmission lines classified
according to the number of conductors they contain, and according to the general class of
electromagnetic wave or propagation ‘mode’ that they support.
The first example (Figure 3.1(1)) has no conductors at all – it is just a rod of dielectric, but
it can still trap and guide an electromagnetic wave. This is extremely important practically
in the form of an optical fibre. It can also be used at ‘high’ radio frequencies, i.e. in micro-
wave or millimetre-wave bands, when it would be referred to as a ‘dielectric waveguide’.
There is no very obvious way we could apply concepts like voltage and current to this
structure.
In Figure 3.1(2) we have a transmission line with only one conductor – a conventional
rectangular waveguide. In Figure 3.1(3) we see a more modern ‘finline’ or ‘E-plane’ structure.
Here there is a central section with a printed conductor pattern, lending itself to the produc-
tion of a microwave integrated circuit. This is considered an attractive structure for work
at millimetric frequencies. Notice that these still do not much resemble the simple idea of
Signal Transmission, Network Methods and Impedance Matching 93

metal dielectric
NO CONDUCTORS MODE CLASS
1. Dielectric waveguide (optical fibre) Non-TEM

ONE CONDUCTOR:
2. Metal waveguide Non-TEM

printed conductor pattern

3. Finline Non-TEM

TWO CONDUCTORS:
(a) TEM TYPE
4. Coaxial cable TEM

5. Parallel wires TEM

ground plane
6. Stripline TEM
ground plane
‘live’ conductor, printed pattern
(b) QUASI-TEM TYPE ‘live’ conductor, printed pattern

7. Microstrip Quasi-TEM
ground plane
‘live’ conductor
ground plane ground plane
8. Coplanar waveguide Quasi-TEM

9. Coplanar strips Quasi-TEM

printed conductor pattern


ground
10. Suspended substrate stripline Quasi-TEM

11. Parallel wires in dielectric support

(c) AN EXCEPTION slot

12. Slotline Non-TEM

Figure 3.1 Classes of transmission line


94 Microwave Devices, Circuits and Subsystems for Communications Engineering

a circuit – they are both short-circuit at DC – which is connected with their description as
non-TEM (transverse electromagnetic) structures. All the remaining transmission line examples
have been classified as two conductor lines. (Sometimes more than two conductors are used
but they still fall into the same general type.)
All transmission lines have some tendency to radiate into space the power they are meant
to be guiding, especially at bends and discontinuities. The enclosed structure of the coaxial
cable, Figure 3.1(4), largely prevents this and makes it suitable as a general-purpose radio
frequency line. The parallel wire line, Figure 3.1(5), may be seen in old-fashioned open
telephone lines, overhead power lines, and sometimes as lines connecting high-power, low
and medium frequency radio transmitters to their antennas.
In Figure 3.1(6) we have a structure suitable for microwave integrated circuits, where the
‘live’ conductor may be given a complex pattern by printed circuit methods. However, it is
mechanically awkward to include other electronic components in it and to assemble.
The structures in Figures 3.1(7), (8), (9) and (10) retain the suitability for ‘printed’ pro-
duction methods and microwave integrated circuits (MICs) while avoiding the mechanical
drawbacks of Figure 3.1(6). Microstrip, Figure 3.1(7), is by far the most widely used, while
coplanar waveguide, Figure 3.1(8), is gaining in popularity. The line in Figure 3.1(10) is
especially useful in low-loss applications such as filters. The structure shown in Figure 3.1(9)
is used only for a few special purposes.
Note that, in the microstrip form of line, it is easy to break the ‘live’ conductor in order to
insert a component in series with it, but if we want to connect a component in shunt between
the live conductor and ground, we have to cut or drill the dielectric. Coplanar waveguide and
coplanar strips do not suffer from this problem.
The line in Figure 3.1(11) is a variant of the parallel wire line where the mechanical
support is built in. It is mainly used for relatively short runs linking radio equipment and
antennas at VHF frequencies. The slotline, Figure 3.1(12), can be, and is, used for complex
MICs but it remains rather specialised and is not particularly easy to use.

3.2.2 Mode Classes


The following important points can be made about the classification of transmission lines:

1. All the two-conductor lines (except slotline), and only these, are classified as transverse
electromagnetic (TEM), or quasi-TEM mode, lines.
2. The lines in this class can be recognised as those that could carry DC excitation and
which conform to the idea of a complete circuit with ‘go’ and ‘return’ conductors.
3. In the two-conductor family, TEM lines can be recognised as those in which the dielec-
tric constant is uniform over the cross-section of the line, while those with a non-uniform
dielectric are quasi-TEM lines.

(In a few special cases, magnetic materials may also be involved, and here the permeability
also has to be uniform for a true TEM line.) A further important point is that:

4. All the TEM and quasi-TEM lines can treated, to a good first approximation at least, by
distributed circuit theory.
Signal Transmission, Network Methods and Impedance Matching 95

Slotline, Figure 3.1(12), looks as though it should be classed as a quasi-TEM line, and it
would support DC excitation. It turns out, however, to be a special case that is not adequately
described by quasi-TEM mode theory. (This is because the conductors are nominally infinite
in extent. Anticipating something to be discussed later, the magnetic field lines in the slotline
mode cannot form complete loops in a transverse plane, because they would have to penetrate
the conductors to do so.) This and the other non-TEM lines need more difficult field theory
treatments that are beyond the scope of this text.

3.3 The Two-Conductor Transmission Line: Revision of


Distributed Circuit Theory
The physical reality of the waves on a transmission line involves fields continuously dis-
tributed in the space between the conductors, and a current density continuously distributed
over the conductors. Distributed circuit theory is a simpler way of analysing the line. It
has the advantage of avoiding field theory, and using circuit concepts, that are easier to
understand. It is also useful in providing insight into the quasi-TEM lines for which the
exact theory is rather difficult.
Distributed circuit theory makes a number of assumptions, namely:

1. The voltage, v, across the line is a well-defined quantity at any transverse plane along its
length
2. The distributed parameters:
Inductance L
Capacitance C
Resistance R
Conductance G
are all specified per unit length and assumed to be well defined and to make sense
physically.
3. The current density can be integrated over a conductor to find the total, or ‘lumped’,
longitudinal current, i, at any point on the line. It is assumed that v and i, together with
the distributed parameters, provide a sufficient basis for analysis without worrying about
the detailed structure of the fields.

In treating junctions between different lines, or between lines and other components, we
usually also assume that:

4. Kirchhoff’s laws are obeyed at junctions; this is a good approximation if the transverse
line dimensions are electrically small, i.e. much less than a wavelength. (Longitudinal
dimensions can be as large as one wants.)

It is important to realise that these are only assumptions, even though they may at first sight
seem self-evident. (1), (2) and (3) can, in fact, be shown to be exactly justifiable for ideal
TEM lines, while for quasi-TEM lines they are only approximations, valid for moderate
frequencies, but nevertheless very useful.
96 Microwave Devices, Circuits and Subsystems for Communications Engineering

Parameter C, which has units of farad per metre (F/m), can be worked out as just the
capacitance found when DC is applied across the line. L, R, and G have units of henry per
metre (H/m), ohm per metre (Ω/m) and siemen per metre (S/m), respectively.

3.3.1 The Differential Equations and Wave Solutions


Having made the assumptions stated in Section 3.3, we now take an infinitesimal section of
the line with length δx, draw an equivalent circuit for it as shown in Figure 3.2, and analyse
it using normal circuit concepts.
The voltage, v, and current, i, on the line are functions of both position and time:

v = v(x,t) (3.1)

i = i(x,t) (3.2)

The voltage across the infinitesimal section of line (with positive on the left) is −(∂v/∂x)δ x
and this can be equated to the voltage developed across the series resistance and inductance.
The current leaving the section on the right is smaller than that entering on the left by an
amount −(∂i/∂x)δ x, and this must be equated to the current flowing through the shunt
conductance and shunt capacitance. The following two equations can thus be deduced:

∂v ∂i
δ x = − ( Rδ x )i − ( Lδ x ) (3.3)
∂x ∂t

Figure 3.2 Lumped parameter model of an elemental length of transmission line


Signal Transmission, Network Methods and Impedance Matching 97

∂i ∂v
δ x = − (Gδ x )v − (Cδ x ) (3.4)
∂x ∂t

Dividing through by δx:

∂v ∂i
= − Ri − L (3.5)
∂x ∂t

∂i ∂v
= − Gv − C (3.6)
∂x ∂t

To progress further with these equations, we have to consider a single harmonic (i.e. sinu-
soidally time varying) component of i and v. For such a single frequency component we can
write:

v = Ve jω t (3.7)

i = Ie jω t (3.8)

Note that we are now using the exponential phasor notational convention of dropping the
‘real part’ sign. V and I are now complex but are dependent on position only, i.e.:

V = V(x) (3.9)

I = I(x) (3.10)

Differentiating with respect to x and t:

∂v dV ( x ) jω t
= e (3.11)
∂x dx
∂i dI( x ) jω t
= e (3.12)
∂x dx
∂v
= jω V ( x )e jω t (3.13)
∂t

∂i
= jω I( x )e jω t (3.14)
∂t

Substituting Equations (3.11)–(3.14) into Equations (3.5)–(3.8) we have:

dV jω t
e = −RIe jω t − Ljω Ie jω t (3.15)
dx

dI jω t
e = −GVe jω t − Cjω Ve jω t (3.16)
dx
98 Microwave Devices, Circuits and Subsystems for Communications Engineering

Finally, we can divide through by e jω t to obtain two equations that are vital to understanding
wave propagation on the line:

dV
= −(R + jω L)I = −ZI (3.17)
dx

dI
= −(G + jω C)V = −YV (3.18)
dx

Z is the series impedance per unit length, and Y the shunt admittance per unit length.
Differentiating with respect to x:

d 2V dI
2
= − ( R + jω L) = ( R + jω L)(G + jω C)V (3.19)
dx dx

d2I dV
2
= − (G + jω C) = ( R + jω L)(G + jω C) I (3.20)
dx dx

Each of these is a standard differential equation, the solution for V being of the form:

V(x) = V1e−γ x + V2eγ x (3.21)

where V1, V2 and γ are suitable constants. It can be easily verified by direct substitution that:

γ = ( R + jω L)(G + jω C) = ZY (3.22)

γ is the complex propagation constant, and is commonly written as γ = α + jβ where α is the


attenuation constant (in neper m−1) and β is the phase constant (in radian m−1).

Self-assessment Problems
3.1 What is d/dx of the function Veγ x, where V, γ are constants? Now differentiate again
and substitute to show that a function of this form can satisfy the equation in
d 2 V/dx 2, given the stated value for γ. Why does the solution still work if we
replace γ by −γ ?

3.2 If a quantity changes by 1.0 neper (−1.0 neper), it has increased by a factor of e
(decreased by the factor 1/e). Prove that a 1.0 neper change of voltage is equivalent
to 8.686 decibels (to 3 decimal places).

3.3.2 Characteristic Impedance


Differentiating Equation (3.21) with respect to x and equating to Equation (3.17):

dV
= − γ V1e−γ x + γ V2 eγ x = −ZI (3.23)
dx
Signal Transmission, Network Methods and Impedance Matching 99

Therefore:

γ γ
I( x) = V1e −γ x − V2eγ x (3.24)
Z Z

Now, from Equation (3.22):

γ ZY Y
= = (3.25)
Z Z Z

and writing out Y and Z in terms of the parameters G, C, R, and L we have:

γ G + jω C
= (3.26)
Z R + jω L

The reciprocal of this quantity is called the characteristic impedance of the line, usually
written Z0, i.e.:

V1e −γ x − V2eγ x
I( x) = (3.27)
Z0

V1 and V2 are constants (independent of x) determined by boundary conditions, i.e. by the


terminating impedance and voltage source at the end(s) of the line. They have dimensions
of volts.
V1 is the complex coefficient of a forward wave, referenced to the specified origin x = 0,
and V2 is the coefficient of a reverse wave. The characteristic impedance Z0 is not the ratio of
total voltage over total current on the line at any point. The reason for this is the minus sign
that appears in Equation (3.27). This equation shows, as we would expect, that the current
is reversed for the reverse travelling wave. The characteristic impedance can, therefore,
be defined as the ratio of voltage over current for either the forward wave or reverse wave
considered alone (remembering, of course, the change of current sign for the reverse wave).
Thus we have:

R + jω L
Z0 = (3.28)
G + jω C

3.4 Loss, Dispersion, Phase and Group Velocity


We have mathematical expressions for the forward and reverse travelling waves, but it is
important to be able to visualise what these mean in terms of waveforms in space and time.
We have:

v(x,t) = V1e jω te−γ x + V2e jω teγ x (3.29)

V1e jω te −γ x − V2e jω teγ x


i( x , t ) = (3.30)
Z0
100 Microwave Devices, Circuits and Subsystems for Communications Engineering

Alternatively, rearranging exponential factors:

v(x,t) = V1e−α xe j(ω t−β x) + V2eα xe j(ω t+β x) (3.31)

V1e −α xe j (ω t−β x ) − V2eα xe j (ω t+ βx )


i( x , t ) = (3.32)
Z0

e j(ω t+βx) in Equations (3.31) and (3.32) represents a wave travelling in the ±x direction and
e+αx represents decreasing (increasing) amplitude with increasing x.
To examine the behaviour of the space-time function Re(e jω te−jβx) let us first put α = 0
and look at the expression for the wave with the negative sign before β. The voltage is
V = Re(V1e jω te−jβx). To keep it simple we can assume V1 is real and take it outside the ‘Re’
sign. (If it isn’t real, the effect is just a simple phase shift anyway.)
Then the voltage is proportional to Re(e jω te−jβx) which can of course also be written, using
the basic properties of exponentials, as Re(e j(ω t−βx) ). We know that this is just cos(ω t − βx).
Imagine a snapshot of this function at time zero – we would just get cos(−βx) and the
graph would be a cosine function drawn along the x-axis.
Now imagine that time advances a little. To get (say) any peak on the cosine function, we
must increase x a bit to get the same value of the argument (ω t − βx). In other words, our
graph of the function in space is moving down the x-axis as time advances.

3.4.1 Phase Velocity


The rate at which the oscillatory pattern appears to move down the x-axis is obviously ω /β.
This is what we call the phase velocity, the velocity at which any single frequency component
appears to be moving in any kind of wave-propagating system. (It will be written vphase or
just vp from now on.)
If, however, we sit at any point in space, i.e. a particular value of x, we just see the
function oscillating sinusoidally in time but of course, the phase of the oscillation is more
and more retarded as we go further down the x-axis. (This is why β is called the phase
constant and is measured in rad m−1.)

Self-assessment Problem
3.3 Prove that ω /β gives the phase velocity: if time increases by δ t, by how much
must x change to keep the argument of the cosine function the same?

3.4.2 Loss
Let us now allow α to be non-zero, so that the voltage will be proportional to Re(e−α xe jω te−jβ x).
The term e−α x is just a real number and we can just take it outside the real-part sign, so this
expression is in fact equal to e−α x cos(ω t − βx).
The term e−α x is an exponentially decaying function of x, and it just multiplies a term that
is the same as the sinusoidal wave travelling to the right which we discussed before.
Signal Transmission, Network Methods and Impedance Matching 101

Figure 3.3 Snapshots of a travelling wave with attenuation

We should therefore picture an exponentially decaying envelope, fixed in space, describing


the amplitude of an oscillation in space and time which travels to the right at the phase
velocity, as pictured previously. The sinusoidal variation is graphed ‘within’ the fixed envelope.
This completes our visualisation of the wave, Figure 3.3.
The reduction of the amplitude of the wave as it travels is called loss or attenuation and it
is obviously due to the absorption of energy in the line. It is clearly the line parameters R and
G that give rise to this loss, and a line that only had inductance and capacitance would be
lossless. In some circumstances, discussed later, it is a good approximation to ignore the loss.

3.4.3 Dispersion
So far only waves at a single frequency have been considered. If a more general waveform
were to be transmitted, it could be decomposed into its frequency components by Fourier
analysis. The waveform could then be observed at a point further down the line, found by
adding up the frequency components again, after their amplitudes and phases have been
modified by the propagation constant of the line.
If the line has a phase velocity that is frequency independent, or at least frequency inde-
pendent over some specified frequency band within which we construct our signal, then it is
said to be dispersionless.
If the line is also lossless over the given band, then all the frequency components travel down
the line ‘in step’ and with no change in amplitude. In this case we would see a time-delayed
102 Microwave Devices, Circuits and Subsystems for Communications Engineering

replica of the original waveform at a point further down the line. The term dispersion refers
to frequency-dependent phase velocity of a transmission structure, or to the distortion of a
transmitted waveform that this produces.
If the line is not lossless and α varies noticeably over the bandwidth of the transmitted
waveform, there is an additional distortion (amplitude distortion) of the waveform due to the
change in relative amplitudes of its frequency components.

3.4.4 Group Velocity


By modulating a continuous wave (CW), i.e. an unmodulated sinusoidal carrier, with a much
more slowly varying modulating signal, we can obtain a narrowband signal (i.e. one whose
bandwidth is much less than its centre frequency). A simple description can be given of the
propagation of such a signal down a dispersive line, which leads to the useful concept of
group velocity.
The simplest example is obtained by multiplying the carrier wave by a sinusoidal modulat-
ing signal at a much lower frequency:
v(t) = cos(ω m t)cos(ωct) (3.33)
If you have previously studied modulation theory, you may recognise this waveform as
that generated if the modulating signal cos ω m t is modulated onto the carrier using double-
sideband, suppressed carrier, amplitude modulation. You should be able to visualise the
waveform, which looks like the carrier wave cos ωc t lying within the amplitude envelope
function | cos ωm t |, and with 180° phase changes of the carrier at the points where cos ωm t
changes sign. You should also know that the spectrum just consists of equal amplitude
sidebands at the carrier frequency plus and minus the modulating frequency.
We shall write this out explicitly using the trigonometric identity:
1 1
cos(a) cos(b) = cos(a − b) + cos(a + b) (3.34)
2 2
which holds for any angles a, b. Equation (3.33) now reads:
1 1
v(t) = cos(ωct − ω m t) + cos(ωct + ω m t)
2 2
(3.35)
1 1
= cos(ω 1 t) + cos(ω 2 t)
2 2
where we have written ωc − ω m = ω 1 for the lower sideband frequency, and ωc + ω m = ω 2 for
the upper sideband frequency.
Now we have two ways of looking at v(t): as a carrier with amplitude modulation, or
simply as the sum of two pure frequencies of equal amplitudes. In the first form we cannot
easily work out how the waveform would travel down a line, but in the second form it is
easy because we know how signals at a single frequency are transmitted.
To keep it simple we shall assume that the voltage is applied to a line that is lossless but
may be dispersive, so that the phase constant will be some known function of frequency,
β = β (ω), but the phase velocity ω /β is not necessarily constant.
Let us write β2 = β (ω 2), β1 = β (ω 1) for the phase constants at the upper and lower sideband
frequencies. It will be assumed that the voltage is forced to be equal to v(t) at the input end
Signal Transmission, Network Methods and Impedance Matching 103

of the line, which we can assume to extend to infinity, or to end in a matched termination, so
that no reflections are present and only waves travelling in the positive direction away from
the input point are excited. At a general point x on the line, the voltages in the lower and
upper frequency travelling waves on the line can now be written respectively as:
1
cos(ω 1t − β1t)
2
and:
1
cos(ω 2 t − β2t)
2
We just have two single frequencies superimposed, and each travelling at its appropriate
phase velocity. Hence the total voltage at a general point on the line is:
1 1
v(x,t) = cos(ω 1 t − β1x) + cos(ω 2 t − β 2 x) (3.36)
2 2
We can rearrange this again using the same trigonometric identity, in its inverse form:

1  1 
cos(c) + cos(d ) = 2 cos  (c + d ) cos  (c − d ) (3.37)
 2   2 
for any angles c, d if we just let: c = a + b, d = a − b. The result is:

1 1  1 1 
v( x, t ) = cos  (ω1 + ω 2 )t − ( β1 + β2 ) x  cos  (ω 2 − ω1)t − ( β2 − β1) x 
2 2  2 2 
(3.38)
 1   1 
= cos ω c t − ( β1 + β2 ) x  cos ω m t − ( β2 − β1) x 
 2   2 
We can now see that the total wave on the line is the product of two terms, each of which
looks like a waveform travelling down the line. One term is the rapidly oscillating carrier
wave at frequency ωc, and the other is the modulation envelope at frequency ωm. If we sit at
one point x on the line and watch the time variations, we again see a single-sideband
suppressed carrier AM signal, but both the carrier oscillations and the modulation envelope
have been shifted in phase, and not necessarily in a way which corresponds to a simple time
delay of the waveform v(t) we started off with at the line input.
On the other hand, we could take, as discussed before, a snapshot of the two terms at a
given time, in which case, each would look sinusoidal in space. At a slightly later time, a
snapshot would show the two sinusoids to have moved in the positive x direction, but not
necessarily by the same amount. The rapidly oscillating carrier wave appears to be moving
at a velocity:
ωc
1
( β1 + β2 )
2
Even if β is varying non-linearly with ω, this becomes asymptotically equal to the phase
velocity ωc /β (ωc) evaluated at the carrier frequency when the modulating frequency tends to
104 Microwave Devices, Circuits and Subsystems for Communications Engineering

zero. On the other hand, the second term, the modulation envelope, appears to be travelling
at a velocity:
ωm
1
( β2 − β1)
2
which is equal to δω /δβ if we write δω = 2ω m = ω 2 − ω1 and δβ = β2 − β1.
In the limit where ωm tends to zero, the velocity of the modulation envelope becomes
equal to the derivative dω /dβ, or 1/(dβ/dω), evaluated at the carrier frequency ωc. This
quantity is called the group velocity, vg:
dω 1
vg = = (3.39)
dβ dβ

Although for simplicity a signal with only two frequency components was considered, it is
clear that the same result would hold for a more general waveform of the form:

v(t) = Re[m(t)e jω t ]
(3.40)
= Re[m(t)] cos ωc t − Im[m(t)] sin ωc t
where m(t) is an arbitrary modulating waveform. (By allowing m to be complex, frequency,
phase or mixed phase/amplitude modulation can be included.) The spectrum of the whole
signal is a shifted image of the spectrum of m, centred on the carrier frequency. If the group
velocity is (to a good enough approximation) constant over the whole spectrum of the
modulated signal, then the modulating signal is transmitted without distortion and travels
down the line at the group velocity.
This condition tells us that a graph of β versus ω is a straight line over the frequency range
of interest. If this straight line passes through the origin, we also have vphase constant and equal
to vgroup over the band in question and the entire signal is transmitted without distortion.
Where the straight line does not pass through the origin, vphase is not constant and not
equal to vgroup. Nevertheless, the modulating waveform is not distorted and the only distortion
of the signal is a progressively increasing phase shift between the modulating signal and
the oscillations of the carrier as we move down the line. Usually this will not matter for a
communications system, or it can easily be corrected at the receiver. A line that is dispersive
in the strict sense, therefore, need not necessarily cause signal distortion that is practically
significant. A more useful criterion may be that dispersion of the modulating signal will only
occur if group velocity is not constant.
A rough criterion is that dispersion will start to cause noticeable distortion of the modula-
tion when:

 d 2β  1 1
x∆ω  2  ≈ radian (3.41)
 dω  8 2
where x is the distance from the starting point and ∆ω is the full bandwidth (in rad/s not Hz)
of the signal being transmitted. The left-hand side of this expression is the approximate
phase error of frequency components at the edges of the band, relative to those near the
carrier frequency, caused by the variation in group velocity.
Signal Transmission, Network Methods and Impedance Matching 105

(You may have come across the almost identical concepts of group delay, group delay
or phase distortion, amplitude distortion and intercept distortion in connection with the
response of linear networks such as filters. The main difference in the present context is that
all these effects increase in proportion to how far we go down the line when observing the
signal. Group delay distortion corresponds to dispersion; intercept distortion corresponds
to the carrier appearing to travel at a phase velocity different from the group velocity of the
modulation envelope.)

3.4.5 Frequency Dependence of Line Parameters


The distributed parameters R, L and G of a transmission line are not in fact constant, but turn
out to be frequency dependent. The variations in R and L are due to the skin effect. Electro-
magnetic waves and their associated currents decay exponentially with depth in a highly
conducting medium, and are greatest at the surface. The skin depth, usually written δ, is the
distance over which a 1/e decay of the E, H fields and current occurs. Where conductivity is
high, as for example in metals, the skin depth is given to a very good approximation by:

2
δ = (3.42)
ω µσ

where σ is the conductivity of the material and µ is its permeability. Notice that skin depth
is inversely proportional to the square root of the frequency. The implication is that at high
frequencies, current is only being carried in a thin layer at the surface of conductors. Since
skin depth is the effective depth over which current is being carried, the resistance of the
conductor increases as the square root of frequency.
Where the transverse dimensions of the conductors in a transmission line are substantially
smaller than the skin depth, the current flow becomes uniform over the conductor cross-
section and the resistance R takes its familiar DC value, which is inversely proportional
to the conductor cross-sectional area. At high frequencies, where the skin depth is small
compared with the conductor dimensions, the resistance increases as the square root of the
frequency, and (for a given line geometry) decreases only in inverse proportion to the linear
size of the conductors, rather than their area (because the current is only flowing in a thin
surface layer). A transition between the two behaviours occurs at frequencies where the skin
depth is comparable with the conductor dimension.

Self-assessment Problems
3.4 For copper, taking σ = 56 × 106 S/m and µ = µ0 = 4π × 10−7 H/m, show that the
skin depth is 9.5 mm at mains frequency (50 Hz) and 0.95 mm at 5 kHz. At what
frequency is the skin depth equal to 1 micron (10−6 m)?

3.5 Show that the DC resistance measured between opposite edges of any square
sheet of a conducting material of conductivity σ and thickness d is 1/σd. Why is
this independent of the size of the sheet? (Hence the term ‘ohms per square’ when
specifying surface resistances.)
106 Microwave Devices, Circuits and Subsystems for Communications Engineering

10 0

10 –1

10 –2

10 –3

10 – 4

10 – 5 0
10 10 2 10 4 10 6 10 8 10 10 10 12
Frequency, Hz

Figure 3.4 Surface resistance of a 1 mm thick copper conductor

Figure 3.4 shows how the surface resistance (measured in ohms per square) of a copper
conductor 1 mm thick varies with frequency. The skin depth becomes equal to the thickness
at 4.5 kHz. Resistance is very close to the value √(ωµ /2σ) over the range 10 kHz to 1 GHz.
At extremely high frequencies, the resistance rises above this value because of the effect
of surface roughness. When the skin depth becomes comparable with the scale size of the
microscopic irregularities on the conductor surface, the current is flowing in a convoluted
(and hence longer) path over these irregularities, and the resistance is increased. In the
example shown, the rms roughness was taken as 1 micron. Because of this effect, care is
often taken to provide a good surface finish on conductors for microwave applications.
The skin effect also causes some frequency dependence of the inductance. Consider, for
example, a total current I flowing in the centre conductor of a coaxial cable. At a radius r
from the centre line, the magnetic field strength H is given by i/2π r, where i is the total
current flowing through a circle of radius r. At zero frequency the current is uniformly
distributed over the conductor cross-section; i, and therefore H, is non-zero right down to
r = 0. Within the space between the conductors, the field is I/2π r.
If, however, the same total current I were flowing near the surface of the inner conductor,
the field outside the conductor would remain as I/2π r but inside the conductor it would
fall to zero within a short distance of the surface. For the same current both the total stored
magnetic field energy, and the total magnetic flux (per unit length of line) between the
centre line and infinity, would be reduced. This argument indicates that, with the skin
effect operating, inductance will be greater at low frequencies and will fall noticeably at the
transition frequency where skin depth becomes comparable with conductor dimensions.
At high frequencies, the magnetic flux within the metal becomes negligible, but the flux in
the space between the conductors remains unchanged; the inductance is therefore expected
to become constant at a rather lower value. Figure 3.5 shows the typical behaviour for a
Signal Transmission, Network Methods and Impedance Matching 107

TEM line with a central conductor diameter of about 2 mm. For typical line geometries L
may be 20–30% greater at low frequencies.
This effect can also be described in terms of the surface impedance of a metal, which can
be shown to be given (to a very good approximation for high conductivity) by:

(1 + j )
Zsurface = (3.43)
σδ

when the conductor is at least two or three skin depths thick. The real part of this expression
is the resistance rising as √ω (already discussed). We see that there is a positive reactive part
of the same magnitude, corresponding to the inductance contributed by the magnetic flux
within the conductor. However, a reactance rising like √ω corresponds to an inductance
proportional to 1/√ω.

Self-assessment Problem
3.6 Explain why this is.

In Figure 3.5 we can see the excess inductance falling with frequency in this manner, above
the transition frequency.
In nearly all everyday RF and microwave engineering, we are operating well above the
skin depth transition frequency of the conductors. Exceptions occur where very thin conductors
are used, for example, in some monolithic microwave integrated circuits (MMICs).

x 10 –7

2.5

1.5

0.5

0 0
10 10 2 10 4 10 6 10 8 10 10 10 12
Frequency, Hz

Figure 3.5 Typical variation of line inductance with frequency


108 Microwave Devices, Circuits and Subsystems for Communications Engineering

For TEM lines, C is constant, and L becomes constant at the high frequencies where the
skin depth is small. In quasi-TEM lines, the distributed circuit method is not an accurate
description at high frequencies; however, constant values of L and C can be assumed in the
frequency range above the skin depth transition frequency but below the frequencies where
the inherently dispersive property of quasi-TEM modes become apparent. (This feature of
the quasi-TEM lines is discussed later.)

3.4.5.1 Frequency dependence of G


Transmission line sections that are deliberately made very lossy may sometimes be used
as attenuators and dummy loads. Loss might be introduced by filling the space between the
conductors with a material that has substantial ohmic conductivity. However, except for
these special cases, our transmission lines are normally filled with dielectrics, such as modern
plastics, that are extremely good insulators. In this case true conduction makes a completely
negligible contribution to G.
However, the parameter G also describes a quite different mechanism of energy absorption,
namely dielectric loss. If an electric field is applied to a dielectric and then removed, not all
the energy put into the dielectric is released again; some is converted into heat (cf. hysteresis
in magnetic or elastic materials). Where the electric field is cycling sinusoidally, there is a
component of the dielectric displacement current (of which more later) in phase with the
applied field, therefore causing power to be dissipated in the dielectric.
The effect is usually quantified by assigning a complex value to the dielectric constant
(relative permittivity) εr of the material; we write:

εr = ε ′r − jε r″ (3.44)

The real part is the dielectric constant in its usual (low frequency) sense, and the ratio of the
imaginary to the real part is conventionally referred to as the loss tangent, written:

ε r″
tan δ = (3.45)
ε r′

(This δ is not to be confused with the δ used for skin depth.) The current flowing into the
line capacitance is jω CV (per unit length) and:

C = (ε ′r − jε r″)Cempty (3.46)

where Cempty is the (conceptual) capacitance of the same line without any dielectric. So we
equate the in-phase component of above current to the term GV, obtaining:

G = ωε r″Cempty (3.47)

εr is always frequency dependent to some extent, but for many dielectrics its real part
is nearly constant, and tan δ is often fairly constant over quite a wide range of frequency.
In a frequency range where ε r″ is constant, we have G directly proportional to frequency,
quite unlike a normal resistor. Materials with very low loss tangents are available for radio
frequency and microwave use. Some examples are given in Table 3.1.
Signal Transmission, Network Methods and Impedance Matching 109

Table 3.1 Lowloss dielectric materials

1 kHz 1 MHz 100 MHz 3 GHz 25 GHz

Alumina ε ′r 8.83 8.80 8.80 8.79 ?


tan δ 0.00057 0.00033 0.00030 0.0010 ?
PTFE ε ′r 2.1 2.1 2.1 2.1 2.08
tan δ < 0.0003 < 0.0002 < 0.0002 < 0.00015 0.0006

3.4.6 High Frequency Operation


Above a certain frequency, a transmission line can be considered to be operating in a high
frequency regime. Under these conditions:

(a) the line has, in a certain sense, less tendency to distort transmitted pulses;
(b) the characteristic impedance becomes real and constant;
(c) a lossless approximation can be made for many purposes.

In the high frequency regime we can also make convenient analytical approximations for α
and β and use them to analyse point (a).
To be operating in the high frequency region, we need the two conditions:

(i) ω L Ⰷ R
(ii) ω C Ⰷ G

Even if the skin effect is operating and R is increasing with frequency, it will only increase
as √ω and the term ω L increases more rapidly with ω . Hence the condition (i) can always
be reached in normal metallic lines. For any given line there is a characteristic transition
frequency at which ω L = R. (Typically this is rather lower than the skin depth transition
frequency, but not by a very large factor.) Condition (ii) looks as though it is also a con-
sequence of making ω large, but this is not true in the commonest case where G is dominated
by dielectric loss and may also be increasing as fast as ω . Rather the condition follows from
using a low-loss dielectric where tan δ is very small.

Self-assessment Problem
3.7 Convince yourself that condition (ii) follows from small tan δ.

If dielectric loss is neglected, the propagation constant can be written as:

R
γ = ( jω L + R) jω C = jω LC 1 + (3.48)
jω L
110 Microwave Devices, Circuits and Subsystems for Communications Engineering

Now assuming R/jω L Ⰶ 1, the second square root can be approximated using the binomial
expansion (or Taylor series) as:

X X2 X 3 5X 4
1+ X ≈1+ − + − +... (3.49)
2 8 16 128

Self-assessment Problem
3.8 Convince yourself of this.

The following approximations can be found for the attenuation and phase constants:

R C
α ≈ + higher order terms (3.50)
2 L

 1 R 
2
5  R 
4
β ≈ ω LC 1 + −  + higher order terms (3.51)
 8  ωL 128  ω L  

Self-assessment Problem
3.9 Set X = R/jω L to obtain the approximations for α and β given in Equations (3.50)
and (3.51).

Equation (3.50) shows that we expect to find the attenuation of transmission lines (in
dB/metre) increasing as the square root of frequency (since R does) most of the time in RF
and microwave work. Although the loss increases with frequency, the rate of increase of
the square root function decreases like 1/√ω , and this implies that the change in attenuation
over a given small band of frequencies falls as the centre frequency rises. Hence there will
tend to be less amplitude distortion of a modulated signal of fixed bandwidth as the carrier
frequency is raised.
The second equation shows that the effect of line resistance on phase velocity is very
small at high frequencies. Also, when R is varying as √ω, the second term in the expression
for β is frequency independent, giving no group delay distortion even though the phase
velocity is not quite constant. Very small group delay distortion arises from the third term.
For an ideal TEM line with R = G = 0 and constant L and C, we would have:

γ = jω Ljω C = jω LC (3.52)

Hence:

β = ω LC (3.53(a))
Signal Transmission, Network Methods and Impedance Matching 111

and:

1
v phase = (3.53(b))
LC
This line would propagate a wave of arbitrary shape with no distortion or loss – we could
break up our arbitrary pulse into its frequency components, which would all travel at the
same velocity (and with no attenuation) and hence would preserve the same pulse shape.

3.4.6.1 Lossless Approximation


Under high frequency conditions, it can be seen from the approximate expressions for α and
β that (even with skin effect operating) lines tend to a ‘lossless’ condition. This does not
mean that the attenuation α itself becomes small, but only the attenuation per wavelength
becomes very small, i.e. the ratio α /β becomes small. Radio frequency circuits such as
filters, matching networks, etc. are generally concerned with wavelength-scale structures and
the lossless approximation is very good when designing them. Lines in this condition also
have Z0 real, to a very good approximation.

Self-assessment Problems
3.10 Ignoring G, show that α /β tends to zero at high frequencies, even with R increasing
as √ω.

3.11 Using Equation (3.28), explain why Z0 becomes real (i.e. purely resistive) when
conditions (i) and (ii) hold. Show that Z0 is then approximately equal to √(L/C).

3.12 Ignoring R and instead letting G be non-zero, and assuming that G is purely
due to dielectric loss, show that α /β is approximately one-half the tan β of the
dielectric.

In almost all RF and microwave work, the conditions (i) and (ii) will be found to hold to a
very good approximation. A real, frequency-independent Z0 can be assumed for TEM lines,
which greatly simplifies design work and also makes it easy to provide a broadband, low
reflection termination (in the form of a resistor) for the transmission lines. For quasi-TEM
lines, Z0 shows weak frequency dependence at frequencies so high that the distributed circuit
model becomes inadequate, but it can be shown to remain resistive.

3.4.6.2 The Telegrapher’s Equation and the Wave Equation


The Telegrapher’s Equation is:

∂2 v ∂v ∂2 v
= RGv + ( RC + LG) + LC 2 (3.54)
∂x 2
∂t ∂t
112 Microwave Devices, Circuits and Subsystems for Communications Engineering

The derivation of this equation implicitly assumes that all the parameters R, G, C and L are
frequency-independent constants (which they may be over wide ranges of frequency). Even
then the equation can only be solved directly for the special case R = G = 0 for which it
reduces to:

∂2 v ∂2 v
= LC 2 (3.55)
∂x 2
∂t

This is called the one-dimensional wave equation. The general solution of the 1-D wave
equation is:

v = f (x − ct) + g(x + ct) (3.56)

where c = 1/√(LC). Note that f and g in this solution are arbitrary functions of a single
variable. (The only requirement is that they are assumed to be ‘smooth’ enough, i.e. to be
twice differentiable.)
The first term in f represents a forward travelling wave, and the second term in g a reverse
travelling wave. c is clearly the velocity of these waves. The implication of this is that a
system obeying the 1-D wave equation can propagate a pulse of arbitrary shape without any
distortion.
When R and G are non-zero, the equation can only be solved by transform techniques,
which means in effect going into the frequency domain and solving at an arbitrary single
frequency. In this case, we can also handle the frequency dependence of R, G, and L, which
of course are not truly constant for real lines.

Self-assessment Problems
3.13 (a) Take ∂/∂x of Equation (3.5) to show that:

∂2 v ∂i ∂ 2i
= −R −L
∂x 2
∂x ∂x∂t

(b) Substitute Equation (3.6) into the second term on the right-hand side of the
above equation.
(c) Now take ∂/∂t of Equation (3.6) and substitute this into the result of step (b).
(Remember that second derivatives are symmetrical, i.e.: ∂ 2i/∂x∂t = ∂ 2i/∂t∂x,
etc.)

You should now have derived the Telegrapher’s Equation.

3.14 Show that for R = G = 0, the Telegrapher’s Equation reduces to Equation (3.55).

3.15 If f (x) is an arbitrary smooth function of a single variable x, and f ′(x) and f ″(x)
represent its first and second derivatives, we can turn f into a function of two
variables x and t by replacing its argument with (x − ct), where c is a constant.
Signal Transmission, Network Methods and Impedance Matching 113

(Note that ∂/∂x of f (x − ct) is f ′(x − ct), and ∂ 2/∂x 2 of f (x − ct) is f ″(x − ct).)
Hence:

(a) Show that ∂/∂t of f (x − ct) is −cf ′(x − ct). (Hint: if not immediately obvious,
this is a simple case of the function of a function rule – or just think about
how much change in the argument (x – ct) is made by changes δ x in x and δ t
in t, respectively.)
(b) Find ∂ 2/∂t 2 of f (x − ct), and hence show that f (x − ct) is a solution of the
1-D wave equation if c = 1/√(LC).
(c) By the same methods show that a function g(x + ct) is also a solution, where
g is a second arbitrary function.

In Self-assessment Problem 3.14 you have proved directly in the time domain, what was
already proved less directly in the frequency domain, that a line with R = G = 0, and
frequency independent constants L, C, can propagate an arbitrary pulse shape without loss
or distortion, at velocity 1√(LC).

3.5 Field Theory Method for Ideal TEM Case


When we first start learning electronics, we become used to thinking in circuit terms where
everything can be analysed in terms of voltages and currents, and Kirchhoff’s laws can be
applied. In RF engineering, and especially at microwave and millimetric frequencies, we have
to start facing the fact that these are only approximations and that a rigorous description
involves us in analysing distributed fields.
Nevertheless we try to go on using circuit theory as far as possible, because it is easier to
understand, is more suitable for analysis and especially synthesis, and can be used in CAD
with limited computing power. Most people find field theory difficult, and in any case many
field problems can only be solved exactly by numerical techniques.
Fortunately most day-to-day RF engineering can still be done on the basis of circuit theory.
In the distributed circuit theory of the line we have done just this, and it works well when
the transverse dimensions of the line are small. Sometimes, for more difficult structures, we
rely on field theory specialists to tackle the field solution and provide us with an equivalent
circuit model which makes it possible to go on using circuit analyses.
A good example occurs in the treatment of discontinuities such as a junction between
two sections of transmission line of different impedance. The circuit view gives us a simple
expression for the reflection. The reality is that there is a complex electromagnetic field struc-
ture set up around the discontinuity, and the reflection coefficient given by the expression
is only an approximation that works well when the transverse dimensions of the structure
remain much less than one wavelength. The situation can be ‘patched up’ by converting the
field solution into an equivalent circuit model of the discontinuity, containing some additional
fictitious circuit elements.
It is not the intention here to give a comprehensive treatment of field theory but it is useful
to give a proper treatment of one special problem where the solution is exact and fairly simple.
This is the TEM wave which, as pointed out in the previous discussion of line classification,
114 Microwave Devices, Circuits and Subsystems for Communications Engineering

exists on a two-conductor line (making the approximation that the conductors are perfect),
where the medium filling the line has properties that are uniform over the line cross-section.
This will give us, at least, some physical insight into the nature of transmission line fields,
and also into how the distributed circuit model can be related to the field picture.

3.5.1 Principles of Electromagnetism: Revision


It is assumed here that the reader has a general knowledge of basic electromagnetics and is
familiar, in particular, with the following principles:

1. Where there is a changing magnetic field linking a loop, there must be an electromotive
force (EMF) induced around the loop. (The law of induction.)
2. Flow of electric current through a loop must be associated with a magnetomotive force
(MMF) acting round the loop. (Ampère’s law.)
3. The flux of electric field out of a closed surface is proportional to the charge contained
within it. (Gauss’s Law.)
4. A changing electric field, even in a vacuum, can act like a current, to be included in law 2.

The third and fourth ideas introduce the idea of displacement current, which may be less
familiar than the first two laws. In a vacuum there is an apparent current density given by
ε0 ∂E/∂t, where ε0 is a fundamental constant of nature called the permittivity of free space
(equal, to four significant figures, to 8.854 picofarad per metre).
The three fundamental laws of electromagnetism, shown above, can be written math-
ematically as follows:

冯 E ⋅ dl = −冮 ∂∂Bt ⋅ ndS
C S
(3.57)

冯 H ⋅ dl = 冮 (J + ∂∂Dt ) ⋅ ndS
C S
(3.58)

冯 D ⋅ ndS = 冮 ρdV = Q
S′
(3.59)
V

In the first two expressions, C denotes any closed loop in space (which need not coincide
with any physical structure), and S denotes any surface that spans that loop. The small circle
on the integral sign emphasises that we are integrating round a closed loop and returning to
the same point at which we started. The dot symbol is for the scalar product of two vectors
and dl denotes an infinitesimal element of C (dl has a direction, so is a vector). dS denotes
an infinitesimal element of area making up S and n denotes the unit normal to the surface S
at any point on it. The requirement that n is normal to the surface does not specify the sign
of n. The right-hand screw rule, with which you may already be familiar, can define the sign
convention. (If the curled fingers of the right hand indicate the sense in which we traverse C,
then the thumb indicates the approximate direction in which n should be pointing.)
Signal Transmission, Network Methods and Impedance Matching 115

Equation (3.57) is usually expressed in words by the statement that the EMF induced
round a loop is equal to minus the rate of change of magnetic flux linking it. Likewise,
Equation (3.58) says that the MMF round a loop is equal to the total current, including dis-
placement current, linking it. In Equation (3.59), S′ is any closed surface and V is the volume
it contains; ρ is the charge density, and Q the total charge, within V.
We are using two quantities, E and D, for electric fields as a convenient way of dealing
with the effect of a dielectric. An electric field acting on a dielectric induces a continuous
distribution of atomic-scale electric dipoles throughout it. We use the symbol P to denote the
total induced dipole moment per unit volume, and define the electric flux density D by the
fundamental relationship:

D = ε 0E + P (3.60)

It can then be shown that the charge Q or charge density ρ only needs to include the con-
tribution from unbalanced total numbers of positive and negative charge carriers, averaged
over the volume of a few molecules, at the point in question. The effect of the atomic dipoles
in the dielectric is automatically taken into account.
When the electric field in the dielectric changes, the movement of charges in the atomic
dipoles represents a real physical current. This current ∂P/∂t is added to the vacuum dis-
placement current to make up the total displacement current ∂D/∂t in Equation (3.58). Then
the current density J only needs to include the conduction current due to the movement of
free charge carriers.
Likewise in magnetic materials there is a distribution of atomic-scale magnetic dipoles
through the material. These are mainly due to electron spin and they behave like currents
circulating in atomic-sized loops. They are conveniently taken care of by using the
relation:

B = µ0H + M (3.61)

where M is the average magnetic dipole moment of the atomic magnets per unit volume,
averaged over the volume of a few molecules, in the material. With this definition the
effect of the atomic magnets is automatically included, and the current J, as stated before,
only has to include the conduction current. µ 0 is a fundamental constant defined exactly as
4π × 10−7 H/m (in the S.I. system of units).

Self-assessment Problem
3.16 Calculate 1/√( µ 0 ε 0) using the values given. What is this quantity?

In the RF context we are usually only dealing with fairly weak fields for which it can be
assumed that P ∝ E and M ∝ B in most materials. We then characterise the medium by
two constants:

B = µH (3.62)

D = εE (3.63)
116 Microwave Devices, Circuits and Subsystems for Communications Engineering

where µ and ε are called the permeability and permittivity of the medium, respectively. It is
usually convenient to write

ε = ε rε 0 (3.64)

µ = µ rµ 0 (3.65)

where ε r, µ r are pure numbers called the relative permittivity and relative permeability. ε r is
also called the ‘dielectric constant’.
For the following treatment of the TEM wave, we need to write the first two electromag-
netic laws in their differential form, viz:

∂B
∇×E=− (3.66)
∂t

∂D
∇ × H = J+ (3.67)
∂t

If you are not familiar with vector analysis, the left-hand side of Equation (3.66) is called the
curl of the field E, and it is a new vector whose (x, y, z) components are given by:

 ∂Ez ∂Ey ∂Ex ∂Ez ∂Ey ∂Ex 


 − , − , −  (3.68)
 ∂y ∂z ∂z ∂x ∂x ∂y 

or equivalently in determinant form as:


x y z
∇ × E = ∂/ ∂x ∂/ ∂y ∂/ ∂z (3.69)
Ex Ey Ez

where x, y, z are unit vectors along the directions of the x, y, z axes.

Self-assessment Problem
3.17 Expand the determinant in Equation (3.69) and show that it gives the same result
as Equation (3.68).

The differential forms of the law are in fact precisely equivalent to the integral forms. The
differential forms can quite easily be deduced from Equations (3.57) and (3.58) by considering
the case where C has been shrunk to an infinitesimal rectangle. (This deduction represents
the proof of Stokes’s theorem.)
For the rest of Section 3.5, we shall assume sinusoidal time variation at a single frequency,
using e jω t notation for this. We also assume the line contains a uniform linear medium with
constants ε and µ. The equations we have to satisfy then reduce to the following (which have
to be satisfied throughout the space between the conductors):
Signal Transmission, Network Methods and Impedance Matching 117

∇ × E = −jωµ H (3.70)

∇ × H = jωε E (3.71)

3.5.2 The TEM Line


The main point that will now be demonstrated is that the time-varying fields in the TEM
wave can be constructed simply from a knowledge of the static electric field. If E t denotes
the static electric field, all we have to do is multiply it by e jω t · e−jβ x, where x denotes distance
along the line and β is a suitably chosen phase constant, to obtain a valid solution of the
Maxwell equations:

E = E t e jω te−jβ x (3.72)

The geometry that will be used is shown in Figure 3.6, for an arbitrary pair of parallel
conductors making up a TEM line. The x-axis is along the line and the y-, and z-axes can be
any pair of axes which, together with x, make up a right-handed coordinate system.

3.5.3 The Static Solution


If we apply a DC voltage to our line, a static electric field is set up on the line. Assuming an
infinitely long line, it is obvious by symmetry that this field is purely transverse, i.e. it has no
x-component. It is equally obvious that the field cannot be dependent on x either, so we shall
write: Estatic = E t ( y, z), where the suffix t means that the field is transverse (no x components)
and the arguments ( y, z) are included as a reminder that the field components are functions
of y and z, but not of x or time. We can write Et in its components as:

(0, Et,y(y, z), Et,z(y, z) )

If we can solve the static problem, i.e. find the functions Et,y(y, z), Et,z(y, z), then we can
calculate the capacitance C per unit length of the line.

Figure 3.6 Coordinate system for TEM line analysis


118 Microwave Devices, Circuits and Subsystems for Communications Engineering

What equations would have to be satisfied for the static field? Because the field is static,
the Maxwell Equation (3.66) tells us that:
∇ × Et = 0 (3.73)
This equation is in fact the condition that, if a test charge is moved from a point A to point
B, the work done is independent of the path, so the potential difference between the two
points is well defined.
We must also assume that any dielectric in the line is perfectly insulating, so that when
we apply the static voltage there is no build-up of charge at any point in the dielectric, and
hence the density ρ of free charges in Equation (3.59) is zero.
Since there is no z field component and the other field components only depend on x
and y, it can be shown from Equation (3.73) that the static field satisfies:

∂Et, z ∂Et ,y
− =0 (3.74)
∂y ∂z

Self-assessment Problem
3.18 Show that Equation (3.74) is correct.

Now consider Equation (3.59) with its right-hand side set to zero. If the volume V is shrunk
to an infinitesimal cuboid, it is quite easy to show that the differential form of the equation
becomes:
∂Dx ∂Dy ∂Dz
+ + =0 (3.75)
∂x ∂y ∂z
In this case D = εE with a constant ε, so we can write:

∂Ex ∂Ey ∂Ez


+ + =0 (3.76)
∂x ∂y ∂z
And finally, E t has no x component (and no dependence on x either!), so the first term in this
expression is zero, leading to:

∂Et, y ∂Et ,z
+ =0 (3.77)
∂y ∂z
In a static field, at any point on a conductor surface, there must be no component of electric
field acting tangentially to the surface. (This holds even if the conductivity is finite.) If any
such field component were present, there would be current flowing along the surface, charges
would be redistributing and the field would not be static. This is called the boundary
condition obeyed by the electric field at a conductor surface.
The same boundary condition is obeyed in a time-varying field if the conductors are
perfect, i.e. of infinite conductivity. Because the conductivity of metals is so high, this can
actually be used as a very good approximation when solving time-varying field problems.
Signal Transmission, Network Methods and Impedance Matching 119

Equations (3.74) and (3.77) together with the boundary condition can be regarded as the
defining equations for the electric field in the static problem. Of course they only define
the form of the field and not its absolute magnitude, which would only be known when the
voltage applied between the conductors has been specified.
Even these equations are not simple to solve for a general conductor configuration, and a
numerical solution has to be used for the general case. Only two cases can be solved with
negligible effort. These are:

(a) Where the lines are wide parallel plates, Et is approximately a uniform, parallel field
normal to the plates – except that at the edges there is a fringing field which is harder to
calculate.
(b) For a pair of coaxial cylinders, the exact solution is that Et is everywhere in a radial
direction, and its intensity is proportional to 1/r where r is the distance from the centre
line.

Analytical solutions can be found, however, by the method of conformal transformations


for a number of useful geometries. The most important of these geometries are a pair
of parallel cylinders, a cylinder parallel to a plane, and a strip conductor parallel to one
ground plane (like a microstrip with no dielectric) or parallel to two ground planes (like
a stripline).
(You may have come across a graphical method known as curvilinear squares that can
be used to find approximately the field lines and equipotential curves for an arbitrary pair of
conductors. With some experience, remembering that the field lines have to enter the surface
normally, you can usually make a fairly good sketch by hand of the approximate shape of
the field lines.)

3.5.4 Validity of the Time Varying Solution


Step 1. Finding the associated magnetic field
Starting from an arbitrary explicit expression for an electric field E, we can work out what
∇ × E is by differentiating it with respect to position. Then if we look at the first of
Maxwell’s Equations, Equation (3.70), we can see that it will tell us what the magnetic field
associated with this electric field would have to be.
This procedure will be used taking the Equation (3.72) as a starting point. This gives:

x y z
∇ × E = ∂/ ∂x ∂ / ∂y ∂ / ∂z (3.78)
0 Et , y ( y, z)e − jβ x Et ,z ( y, z)e − jβ x

(The electric field is assumed to have no x-component.) The first term of the expanded deter-
minant reads: x times [(∂/∂y Et,z · e−jβx ) − (∂/∂z Et,y · e−jβx )]. The term e−jβx does not vary with
y or z so we can take it outside the expression, which reduces to: e−jβx · [∂/∂y Et,z − ∂/∂z Et,y]x.
Now we have already seen that [∂/∂y Et,z − ∂/∂z Et,y] = 0 is one of the defining equations of
the static field, so this first term in the determinant is actually zero.
120 Microwave Devices, Circuits and Subsystems for Communications Engineering

The second term reads: − y times [(∂/∂x Et,z · e−jβ x ) − (∂/∂z of zero)]. Since Et does not depend
on x, the only contribution to this whole term arises from differentiating e−jβ x with respect to
x, which just gives −jβ · e−jβx, so this whole term in the end reads: ( jβ · e−jβx · Et,z) y.
The third term reads: z times [(∂/∂x Et,y · e−jβ x) − (∂/∂y of zero)] and this becomes:
(−jβ e−jβx Et,y)z in exactly the same way. So we have finally shown that:

∇ × E = ( jβe−jβ xEt,z) y − ( jβe−jβ xEt,y)z (3.79)

if E is derived from the static electric field in the way we assumed. Now if this is compared
with the Maxwell Equation (3.70), we can deduce that the associated magnetic field must
be given by:

β
H= [(−e−jβ xEt,z) y + (e−jβxEt,y)z] (3.80)
ωµ

We can now see that the magnetic field has the same components as the electric field,
except that the y and z components are interchanged, the y component is given a minus sign,
and there is an overall coefficient of proportionality β /ωµ. A simple way of expressing this
is that the magnetic field H is also transverse, and it is proportional to the electric field
and at right angles to it. This tells us that the magnetic field lines are at right angles to the
electric field lines!
Another simple way of writing this expression is:

β
H= x×E (3.81)
ωµ

You will be aware that the cross-product is at right angles to both of the vectors making it
up, so this again shows us that the H field is transverse to the line and is at right angles to E.
The operation we just did looks very abstract, but we shall see later that it is fairly easy to
picture physically in terms of the law of induction.

So far we have simply assumed a form for the electric field, and worked out what the
associated magnetic field would have to be. To show that we have a proper field solution, we
need to show that the other Maxwell Equation (3.71):

∇ × H = jωεE (3.82)

can be satisfied. In the process we find what the so far unspecified value for β has to be. We
therefore take the curl of the expression (3.81) to obtain:

x y z
β
∇×H= ∂/ ∂x ∂/ ∂y ∂/ ∂z (3.83)
ωµ
0 − Et, z ( y, z)e − jβ x
Et, y ( y, z)e − jβ x
Signal Transmission, Network Methods and Impedance Matching 121

Remembering that the ∂/∂x and ∂/∂y operate only on the Et terms, and the ∂/∂z only on the
e−jβx we obtain:

β 2 − jβ x β − jβ x  ∂Et,y ∂Et, z 
∇×H= j
ωµ
([ ]
e Et,y )y + (e − jβx Et, z )z +
ωµ
e  ∂y + ∂z  x
 
(3.84)

Again we saw that (∂/∂yEt,y + ∂/∂zEt,z) = 0 is one of the defining equations of the static field,
while the expression [(e−jβ x · Et,y) y + (e−jβ x · Et,z)z] is just the time-varying electric field E again.
So we have shown that:

β2
∇×H= j E (3.85)
ωµ

Then, if we compare this with the Maxwell Equation:

∇ × H = jωεE (3.86)

we see that we have a solution provided that:

β 2 = ω 2µε (3.87)

So we have now shown that, for a transmission line with a uniform dielectric and perfect
conductors, we can generate time-varying electric and magnetic fields that satisfy Maxwell’s
field equations. Both of the fields are purely transverse to the line, and the electric field is
identical in form to the static electric field. The time-varying, travelling wave is consequently
known as the TEM (Transverse Electromagnetic) mode for the line.

3.5.5 Features of the TEM Mode


The TEM mode is convenient in having several features that make its behaviour easy to
analyse, although these are not of great importance to a line’s practical performance:

1. The TEM time varying fields are easy to calculate – it is only necessary to solve for the
static electric field on the line.
2. The voltage across the line is well defined. Because of Equation (3.73), the integral of
electric field from one conductor to the other (keeping within a given transverse plane) is
independent of the path chosen.
3. Because the voltage is well defined, the characteristic impedance defined as voltage over
current in the wave is also well defined. It also agrees with other definitions, e.g. using
the relation that I 2Z0 or V 2/Z0 gives the average power transported in the wave, if I and
V are the rms current and voltage in the wave. The Z0 also turns out to be inherently
frequency independent (see below). For quasi-TEM lines such as microstrip, the various
definitions of Z0 show different, though fairly minor, variations with frequency, while
agreeing at low frequency.
4. Currents on the line are purely longitudinal. On quasi-TEM lines, small transverse
components of current exist.
122 Microwave Devices, Circuits and Subsystems for Communications Engineering

Potentially much more significant in performance terms is the following property:

5. The TEM mode is inherently non-dispersive, i.e. in the ideal case the phase velocity is
independent of frequency, so that a pulse of arbitrary shape can be sent down the line
without distortion. The phase velocity is ω /β which we can now see is given by:

1 c
v phase = = (3.88)
µε µr εr

Two effects can still cause the TEM wave to be slightly dispersive in practice:

(i) The finite conductivity of the conductors, neglected in the ideal TEM theory, has a slight
effect on the phase velocity. As discussed in Section 3.5.3, this becomes negligible at
high frequencies, and more important is the frequency-dependent attenuation that the
resistance produces. (It also produces small components of electric field along the line.)
(ii) For certain materials, ε and µ may be noticeably frequency-dependent. The most
common case is where a non-magnetic ( µr almost 1) and low-loss dielectric is used, and
these materials usually have dielectric constants whose real parts are nearly constant
over very wide ranges of frequency.

However, the dispersion is not an inherent property of the mode itself. This may be con-
trasted with non-TEM modes such as those in a metallic waveguide, which are dispersive
even if it is filled with vacuum and made of perfect conductors. Likewise it will be seen later
that the ‘quasi-TEM’ mode of a microstrip, and surface wave modes on a microstrip substrate,
are inherently dispersive even given perfect conductors and frequency-independent ε r of
the dielectric substrate.
The TEM mode is therefore the best-behaved type for propagating wide-band signals
over long transmission lines with minimum waveform distortion. The quasi-TEM lines are
potentially more distorting, but this is not usually a major issue in RF and microwave sub-
systems where line lengths and signal bandwidths are moderate.

3.5.5.1 A useful relationship


An important and useful relationship can easily be derived from the field theory of the TEM
mode, and even more easily by comparing the field and circuit descriptions. According to
the distributed circuit theory, the velocity of waves on an ideal (lossless) line is 1/√(LC),
while the field theory gives it as 1/√( µε). We can therefore deduce:

µr εr
LC = µε = µr εr µ 0ε 0 = (3.89)
c2

Because perfect conductors have been assumed, there is no field penetration into the con-
ductors and clearly L must be understood as the high frequency limiting value in Figure 3.5,
which is the value that would be calculated using perfect conductors. The relationship is
worth committing to memory, as it is valuable for calculating line inductances, and in the
analysis of quasi-TEM lines.
Signal Transmission, Network Methods and Impedance Matching 123

Self-assessment Problems
3.19 For lossless lines, or lines working at high frequency, use the distributed circuit
theory to deduce the following useful relationships:

Z0 = vp L
Z0 = 1/(vpC)
Z0 = Z0,empty√( µr /εr)

(Subscript ‘empty’ refers to the value that a parameter would take if we could
remove the dielectric from the line while leaving the conductor geometry
unchanged. For the third relation you also need Equations (3.58) and (3.89).)

3.20 Standard coaxial cable has Z0 = 50 Ω (at high frequency). (a) For a cable filled
with polythene dielectric having εr = 2.26, calculate L, C and the phase velocity.
(b) What would be the ‘empty’ values of Z0, C and L for this cable?

3.5.6 Picturing the Wave Physically


The time varying field was derived using the rather abstract-looking differential forms of the
Maxwell Equations, so it is useful to finish this section of the theory with a physical picture
of the wave which shows how the electromagnetic laws are satisfied in terms of the more
familiar circuital laws, and how the theory links up with the distributed circuit description.
Consider Figure 3.7. Looking at loop A in the transverse plane, there is no EMF round
this loop because of the curl-free property of the static E field, and the law of induction is
satisfied because there is no longitudinal component of H and therefore no magnetic flux
through the loop.
If we draw a second longitudinal loop B orientated so that its short edges are normal to
the E field, this loop is normal to the E field at all points so there cannot be any EMF round
it. If it is spanned by a cylindrical surface, the H field, being normal to E, is everywhere
parallel to this surface. The H field lines do not pass through the surface and there is no
magnetic flux through it. Again the induction law is satisfied.
The interesting case is loop C whose edges bc, da lie in the conductor surfaces. There
is no EMF along either edge, because there is (as required by the boundary condition
for perfect conductors) no E field component tangential to the surface. However, the line
integral of E from a to b can be identified as the voltage V(x) across the line at the plane x,
(taking the inner conductor as positive) and that from c to d as minus the voltage at x + δ x.
Hence the EMF round loop C is V(x) − V(x + δx) which is taken as (−∂V/∂x) × δ x when δ x
is small.
You should now be able to see that the magnetic field is flowing through the loop C,
and is in fact at right angles to it. If we took any small sub-loop within C, you should be
able to visualise how the law of induction, Equation (3.57), is satisfied in the TEM wave
by equating the EMF round the sub-loop, which is equal to the area of the sub-loop times
the rate of change with x of the radial electric field, to the rate of change with time of the
124 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 3.7 Form of the TEM wave in a coaxial line

magnetic flux through it. For a sinusoidal travelling wave of the type discussed before,
all the field quantities are proportional to the function cos(ω t − β x). ∂/∂x and ∂/∂t of this
function are both proportional to sin(ω t − βx) so you should be able to picture how the law
of induction can be satisfied in the TEM travelling wave at all times and positions.

Self-assessment Problem
3.21 Prove that ∂/∂x and ∂/∂t of cos(ω t − βx) are both proportional to sin(ω t − βx)
and try visualising it in terms of the snapshots of the cosine function.

Applying the law of induction to the whole of loop C, one of the distributed circuit theory
equations can be deduced. Inductance is defined as the ratio of magnetic flux linking a circuit
to the current flowing in it. The flux Φ linking the loop C is, by definition, L δ x × I where
I is the local value of the current flowing in the centre conductor. (For an ideal TEM
wave we know that the magnetic field is what would be calculated for a static current – but
assuming perfect conductors and therefore no flux penetrating them – and L is frequency
independent and equal to the static value.) If I flows in the positive x direction, the right-
hand screw rule says that the magnetic field is circulating in a clockwise sense when looking
Signal Transmission, Network Methods and Impedance Matching 125

in the positive x direction. However, the right-hand rule applied to loop C says that, in
working out the EMF round C, the flux passing through C has to be defined in the opposite
sense when applying the law of induction. So we can write:
Flux linking C is Φ = −Lδ xI
Therefore:

EMF round C = −dΦ/dt = L δ x ∂I/∂t


But:
EMF round C = (−∂V/∂x) × δ x

as discussed before.
Therefore, cancelling δ x, ∂V/∂x = −L∂I/∂t, which was one of the distributed circuit
equations for an ideal lossless line.

Self-assessment Problem
3.22 Of the three loops shown in Figure 3.7, only one has a non-zero magnetomotive
force round it, and a non-zero electric flux passing through it. Which one is it?
From this starting point, give a similar physical explanation of how the other
circuital law, Equation (3.58), is satisfied at all positions and times.

The other equation of the distributed circuit theory can be deduced by applying Equation
(3.58) to a loop similar to loop B placed close to the surface of the centre conductor, but
it is quicker to do it just using the law of charge conservation. (This law can be shown to
be implicit in Equation (3.58).) Consider a short section of conductor between two trans-
verse planes at x, x + δ x. There is a net current into this piece of conductor given by
I(x) − I(x + δ x), and the charge conservation principle says that this net inflow must be
equated to a rate of build up of charge in that section. Hence we can write:

∂I d
I ( x ) − I ( x + δx ) = δx = (Qδx ) (3.90)
∂x dt
where Q is the total charge per unit length on the conductor, and of course δ x can be cancelled.
At any point on one of the conductors the flux of electric field out of that conductor,
which is ε times the local value of the normal component of E, must be equated to the
surface charge density ρ on the conductor. Knowing the functional form E(x, y) enables us
to integrate E from one conductor to the other to obtain a voltage V, and to integrate ρ round
the conductor circumference in a fixed transverse plane to obtain Q per unit length. The
capacitance per unit length C is defined as the ratio Q/V. The circuit approach reduces the
complexity of the field function to a ‘lumped’ pair of variables Q and V and the relation
Q = CV. We now have that ∂I/∂x = −dQ/dt = −C∂V/∂t, which is the second distributed
circuit equation. In the TEM time varying wave the E field has the same form as the static
field, and so the capacitance C is frequency independent and takes its static value.
126 Microwave Devices, Circuits and Subsystems for Communications Engineering

Again picturing the sinusoidally varying travelling wave cos(ω t − β x), you should be able
to visualise the charge density growing with time at the points on the centre conductor where
the spatial gradient of I is negative, the electric field correspondingly growing with time at
those points, and the space and time rates of change both proportional to sin(ω t − β x) as
before.
The time varying field was found as a solution of Maxwell’s Equations in the space between
the conductors, together with the boundary condition at the conductor. In this approach the
fields tell us what the charges have to be, rather than the other way round. This is fine because,
in assuming that boundary condition, we are assuming that the conductors are perfect and
that the charges can redistribute themselves instantaneously to produce the given fields.
At the end of all this work we have used the rigorous field theory to justify the distributed
circuit view of the line as an exact description when the line is ideal (lossless) and of true
TEM type.
It is easy to see that the pure TEM mode cannot exist on one of the lines where the
dielectric constant is not uniform over the cross-section, as the wave would have to take
two different velocities in the dielectric and air regions. However, it can be shown that, at
low frequencies, a ‘quasi-static’ approximation can be made. In this regime, the longitudinal
components of the E and H fields are small, and the transverse parts approximate closely
to their static form. (The static form of H would have to be calculated assuming perfect
conductors, i.e. no penetration of H into the conductors and no normal component of H
at the surface.) This field pattern is called the ‘quasi-TEM’ mode of the line. The distributed
circuit description of the line continues to work well in this regime.

3.6 Microstrip
This transmission technology has been important for at least two and a half decades already,
and microstrip technology remains at the forefront of the options for RF, microwave and high-
speed systems implementation. Its significance is actually increasing because of the expanding
applications for RF and microwave technology as well as high-speed digital electronics.
Currently many companies are appreciating that microstrip technology is the best answer
to problems associated with maintaining the operating integrity of digital systems clocked at
frequencies around and above 200 MHz. Examples of systems using microstrip include:

1. Satellite (DBS, GPS, Intelsat, VSATs, LEO systems, etc.).


2. Wireless (cellular, PCN, WLAN, etc.).
3. EW (ECM, ECCM), radars and communications for defence.
4. High-speed digital processors.
5. Terminals for fibre optic transmission.

Many other examples could be cited.


Microstrip as a concept begins with the requirement for increased integration at RF and
microwave frequencies, leading to microwave integrated circuits (MICs). This requirement
was first recognised in the late 1960s, mainly driven by military applications and hence
leading to designs such as broadband EW amplifiers, etc. Although this class of applications
is still growing, in spite of overall defence reductions, it is the other applications sectors
listed above that drive most current interests in microstrip technology and design.
Signal Transmission, Network Methods and Impedance Matching 127

(a) Completely non-TEM (b) Quasi-TEM

ε0 ε ε0
ε
(i) Image line (ii) Microstrip

ε0
ε ε0 ε

(iii) Finline (iv) Coplanar waveguide (CPW)

ε0
ε ε
(v) Slotline ε0
(vi) Inverted microstrip

ε0
ε

(vii) Trapped inverted microstrip (TIM)

ε0
ε

(viii) Suspended stripline

Figure 3.8 A range of planar and semi-planar transmission line structures


suitable as candidates for microwave circuit integration.
(Substrates are shown shaded in each case and metal is shown in dense black.)

Until the late 1960s the great majority of RF and microwave transmission used these
types of structures – and they are still used extensively. However, in order to accommodate
both active and passive chip insertion, some form of planar transmission structure is required.
Conventional PCBs (even single-layer ones) are unsatisfactory, mainly because of their
radiation losses and cross-talk problems. Multilayer PCBs are worse in these respects.
Microstrip is totally grounded (earthed) on one side of the dielectric support, which pro-
vides a better electromagnetic environment than open structures (such as PCB conductor
tracks).
A range of possible MIC-oriented transmission line structures is shown (as cross-sections)
in Figure 3.8. In each case ε0 indicates ‘air’ (strictly vacuum) and εr is the relative permittivity
of the supporting dielectric. Here we always refer to this supporting dielectric as the ‘substrate’
(with thickness denoted by h).
Of the structures shown in Figure 3.8 the most important are: microstrip, CPW, finline and
suspended stripline. The rest of this chapter, however, is entirely concerned with microstrip.
128 Microwave Devices, Circuits and Subsystems for Communications Engineering

air
ε = ε0
Top conducting
strip
t
Dielectric w
substrate y
ε = ε 0 εr h
x

Ground plane
(conducting)

Figure 3.9 Three-dimensional view of microstrip geometry

3.6.1 Quasi-TEM Mode and Quasi-Static Parameters


As a rough first approximation, electromagnetic wave propagation in microstrip can be
likened to that in coaxial lines, i.e. transverse electromagnetic (TEM). This, of course,
implies that all electric fields should be transverse to, and orthogonal with, the magnetic
fields.
However, even a cursory inspection of the structure (see Figure 3.9) reveals a distinct
dielectric discontinuity – between the substrate and the air – and any mixed medium like this
cannot support a true TEM mode – or indeed any ‘pure’ mode (TE, TM, etc.).
Because the presence of the ground plane (the grounded ‘underside’ of the structure)
together with the substrate concentrates the field between the strip and the ground there is
some reasonable resemblance to a ‘flattened-out’ coaxial (‘co-planar’) line. It is this feature
that gives rise to the ‘quasi-TEM’ concept and suites of design approaches (curves and
formulas) have been developed that provide engineering design level accuracy, at relatively
low frequencies at least.

Self-assessment Problems
3.23 Why cannot microstrip support any pure (TEM, TE, TM) mode?

3.24 Give two reasons why microstrip is preferred above other media for MICs.

3.6.1.1 Fields and static TEM design parameters


A simplified cross-section showing only the electric field is shown in Figure 3.10, with
magnetic and electric fields shown in three-dimensional detail in Figure 3.11(a) and (b).
The presence of longitudinal components of fields is clear from these diagrams, and in the
next subsection we show the importance of quantifying the effects of these for accurate
high-frequency design.
Signal Transmission, Network Methods and Impedance Matching 129

Figure 3.10 Simplified cross-section, showing electric field only

(a) Magnetic field distribution

(b) Electric field distribution (partial view)

Figure 3.11 Three-dimensional views showing (a) magnetic field alone;


(b) electric field alone (in air, only for simplicity)

3.6.1.2 Design aims


Comprehensive passive networks can be designed and built using interconnections of micro-
strips. Given the relative permittivity εr and the thickness h (usually in mm) of the substrate,
as well as the desired characteristic impedance Z0 (in ohms) and the frequency of operation
f (in GHz), the design aims are to determine:

1. the physical width w of the strip, and


2. the physical length l of the microstrip line.

Initially, the static TEM parameters can be used to find w but first we need to define two
special quantities relating to microstrip, namely, the effective microstrip relative permittivity,
εeff, and the filling factor, q.
130 Microwave Devices, Circuits and Subsystems for Communications Engineering

The quantity εeff simply takes into account the fact that propagation is partly in the sub-
strate and partly in the air above the substrate surfaces. The maximum asymptotic value of
εeff is εr and the lowest possible asymptotic value is 1.0 (for air).
The filling factor, q, also takes into account the fact that propagation is partly in the
substrate and partly in the air. However, q just tells us the proportionate filling effect due to
the substrate, and its maximum possible asymptotic value is 1.0 (representing fully filled),
with a lowest possible asymptotic value of 0.5 (implying an exactly evenly filled space).
Both εeff and q must be used together in static TEM design synthesis.
It can be shown that the following useful relation holds:

εeff = 1 + q(εr − 1) (3.91)

Also the characteristic impedance of the structure in free-space Z01 (i.e. entirely air-filled),
which is a useful normalising parameter for design purposes, is given by:

Z01 = Z0 ε eff (3.92)

Neglecting dispersion (see Section 3.6.2), the wavelength (λ g) in a line intended to be one
wavelength long is given (in millimetres), for frequency F (in GHz), by:

300
λg = mm (3.93)
F ε eff

Self-assessment Problem
3.25 Calculate the physical length of a three-quarter-wavelength microstrip when εr is
9.8, the filling factor is 0.76 and F = 10 GHz.

3.6.1.3 Calculation of microstrip physical width

In practice, microstrip physical width, w, is almost always determined within computer CAD
(CAE) routines. We can use graphical approximations for w (as well as other parameters),
however, and one approach is outlined here. Such methods provide excellent checks on
computed results.
Presser (following the fundamental work of Wheeler) developed the data for the curves
shown in Figure 3.12.
We now run through the sequence of steps required to use these curves, with the follow-
ing warning:

It will frequently be found that designs require extrapolation of the curves beyond their
useful accuracy. These curves should only be used when initial calculations indicate that
mid-ranges are applicable.
Signal Transmission, Network Methods and Impedance Matching 131

Microstrip filling fraction (q)


1.0 0.9 0.8 0.7 0.6 0.5
10
9
8
7
6
5
4
q
3
Microstrip shape ratio w/h

z01

1
0.9
0.8
0.7
0.6
0.5
0.4

0.3

0.2

0.1
0 50 100 150 200 250
Air-spaced ‘microstrip’ characteristic impedance (z01) ohms

Figure 3.12 Generalised curves facilitating approximate analysis or synthesis of microstrip

The calculation sequence is:

1. Make the initial (very approximate) assumption the εeff = εr. This only provides starting
values. (To accelerate the process you can take εeff just below the value of εr.)
2. Calculate the air-spaced characteristic impedance under the approximation of step 1.
3. Use Figure 3.13 to find the width to height ratio, w/h, applicable to this value of airspaced
characteristic impedance. Also note the corresponding value of q.
4. Calculate the updated value of q using Equation (3.91).

This completes one iteration of the sequence. In most cases at least three iterations should be
completed before final values are established.
132 Microwave Devices, Circuits and Subsystems for Communications Engineering

Here we shall confine our discussion to microstrips on uniform, isotropic, and non-ferrite
substrates. (Occasionally exotic substrate microstrips are required, design guidance for which
is available in the specialist literature.)
Important substrates include: alumina, co-fired ceramics, plastic (e.g. PTFE, glass-
reinforced) and semi-insulating semiconductors such as GaAs, InP and, of course, silicon.
At moderate to high microwave frequencies GaAs is often used – particularly where medium
to high volumes are required.

Self-assessment Problem
3.26 Using the same substrate as in SAP 3.25, and desiring a characteristic impedance
of 31.3 ohms, determine, using graphical synthesis, the width of the microstrip if
the substrate thickness is 0.5 mm.

Relatively accurate formulas for analysing and synthesising microstrip are available.
Design-orientated texts provide several suitable expressions and give guidance on limits of
applicability.

3.6.2 Dispersion and its Accommodation in Design Approaches


In any system, non-linearity in the frequency f versus wave number β (or phase coefficient)
results in what is termed dispersion. One manifestation of the presence of such dispersion is
group-delay distortion of signals transmitted through such a system.
Dispersion can be chromatic, ‘waveguide’, or modal. Chromatic dispersion is observed
in many materials and extends through optical as well as microwave and millimetre-wave
frequencies. In fact, all the varieties of dispersion listed are important in optical fibres to
varying extents.
All of the planar and semi-planar microwave transmission structures are dispersive and
microstrip is no exception. In this case the dispersion is modal and arises from variations in
coupling between longitudinal section electric (LSE) and longitudinal section magnetic (LSM)
modes. That these types of modes must exist should be clear by re-visiting Figure 3.11. As
the frequency is increased so the strength of coupling between modes also increases and,
because the currents associated with most of the modes are concentrated beneath the strip
(adjacent to the dielectric), the fields overall are proportionately contained more within the
substrate. Hence, the effective microstrip permittivity increases. For design purposes the
questions that must be answered are:

1. Precisely how much is this increase?


2. How does this vary with other microstrip parameters?

Since the effective microstrip permittivity is known to be frequency ( f ) dependent, we


acknowledge this fact writing it as εeff ( f ). This represents a very important microstrip design
parameter – especially at high frequencies.
Signal Transmission, Network Methods and Impedance Matching 133

εr
f ∞

EFFECTIVE
MICROSTRIP εeff(f)
PERMITTIVITY
εeff(f)
εeff
0 f

FREQUENCY f

Figure 3.13 Microstrip dispersion: εeff ( f ) versus frequency f

Calculations of wavelength in microstrip must now be based upon εeff ( f ) and not (at high
frequencies at least) using the low frequency parameter εeff . The formula for wavelength
therefore becomes:

c
λg = (3.94)
f ε eff ( f )

Calculations of w are substantially unaffected by this dispersion (although strictly the


characteristic impedance is also frequency dependent). In most cases w can be computed as
was shown earlier.
Dispersion is therefore generally viewed in terms of varying εeff ( f ) the limits being
clearly defined in Figure 3.13.
As might be expected, early attempts to fully quantify this dispersion were based upon
various frequency-dependent field solutions (founded ultimately upon Maxwell’s Equations).
One of the earliest offerings was the work of Itoh and Mittra who employed a spectral
domain method to solve this problem.
A clear difficulty with all these approaches, however accurate, is the extensive computa-
tional time required to find even εeff ( f ) and hence wavelength in the microstrip. Also, even
for relatively narrow bands of operation, the wavelengths are generally required over a range
of frequencies and for as many microstrips as there are in the final design. Consequently,
a search has been conducted over many years, for practical closed form expressions (or
families of expressions) that would enable microstrip dispersion to be quantified rapidly
and efficiently. A major breakthrough was achieved in 1972 by Getsinger, who formulated
and published his ‘microstrip dispersion model’. This was proven to work accurately, within
around 3% or so, for microstrips on alumina-type substrates operating up to about 12 GHz.
Getsinger modelled microstrip line as separated structures for which approximate analysis
was more straightforward and provided the following expressions:
134 Microwave Devices, Circuits and Subsystems for Communications Engineering

ε r − ε eff
ε eff ( f ) = (3.95)
1 − G( f / f p )2

where:

Z0
fp = (3.96)
2µ 0h

and µ 0 is the free-space permeability (4π × 10−7 H/m). The parameter G is purely empirical,
thereby giving some flexibility to the formula. G is dependent mainly upon Z0 but also to a
lesser extent upon h. Getsinger deduced from measurements of microstrip ring resonators on
alumina that:

G = 0.6 + 0.009Z0 (3.97)

These expressions are accurate (typically to within about 3%) where alumina substrates having
thickness between about 0.5 and 1 mm are used and frequencies remain below 12 GHz.
For other substrates, including plastics with much lower permittivities, the formulas are still
reasonably accurate up to this frequency – although Z0 should remain above 20 ohms. Where
substrates differing from plastics or alumina are used, and particularly when the frequency
rises above 12 GHz or so, different expressions are necessary.
Since Getsinger’s original work, many variations have been devised and reported. One
example is that due to Kobayashi, who developed the following linked relationships:

ε r − ε eff
ε eff ( f ) = ε r − (3.98)
1 + ( f/f50 )m

where:

fk , TM0
f50 = (3.99)
0.75 + (0.75 − (0.332 /ε 1r.73 ))w / h

 ε eff − 1 
c tan −1  ε r
 ε r − ε eff 
fk , TM0 = (3.100)
2π h ε r − ε eff

m = m0 mc (3.101)
3
1  1 
m0 = 1 + + 0.32   (3.102)
1 + w/h  1 + w/h 

 1.4   − 0.45 f  
1 + 0.15 − 0.235 exp   , where w / h ≤ 0.7 (3.103)
mc =  1 + w/h   f50  
1, where w / h > 0.7

Signal Transmission, Network Methods and Impedance Matching 135

10
Effective microstrip permittivity

9.8

9.6

9.4
Kirschning & Jansen
Yamashita
9.2
Kobayashi

9.0
0 10 20 30 40
Frequency (GHz)

Figure 3.14 Dispersion curves for a microstrip on a GaAs substrate (comparison of predictions)

The consistency of results computed using various closed form dispersion expressions is
indicated by the curves in Figure 3.14, for microstrip on a 0.127 mm thick GaAs substrate
(εr = 13) and microstrip width (w) of 0.254 mm.
It is noteworthy that the closest agreement is between the curves of Kirschning and Jensen
and Kobayashi and that the curves extend to 40 GHz.

Self-assessment Problem
3.27 Using Getsinger’s expressions, calculate, for the same microstrip line on alumina as
applies to previous SAPs, the wavelength at 11.5 GHz. Re-calculate the wavelength
for a microstrip with an impedance of 80 ohms (all other parameters identical).
What does this tell us about the effect of impedance on the ‘degree of dispersion’?

3.6.3 Frequency Limitations: Surface Waves and Transverse Resonance


The generation of surface waves, comprising electromagnetic energy trapped close to the
surface of the substrate, and also a transverse resonance effect, both set limits to the
maximum operating frequency associated with a particular microstrip.
The lowest-order TM surface wave mode is the first limiting phenomenon in this category
and analysis leading to the corresponding frequency limit starts by considering the substrate
as a dielectric slab. An eigenvalue expression is set down and the net phase coefficient is
calculated, from which the relationship for the TM surface wave frequency is determined as:

c tan −1ε r
fTEM1 = (3.104)
2π h ε r − 1
136 Microwave Devices, Circuits and Subsystems for Communications Engineering

For narrow microstrips on reasonably high permittivity substrates (usually greater than
εr = 10), which is typically the most critical situation, this reduces to:

106
fTEM1 = (3.105)
h εr − 1

leading to the following formula for maximum substrate thickness allowable while avoiding
the generation of this wave:

0.354λ 0
h= (3.106)
εr − 1

Clearly, keeping the substrate as thin as possible (consistent with other engineering constraints)
ensures that the shortest possible wavelength, i.e. the highest frequency, is supported while
avoiding the generation of the lowest order TM surface wave.
We next consider the lowest order transverse microstrip resonance. A resonant mode
can become set up transversely across the width of a microstrip, which can couple strongly
with the dominant quasi-TEM mode. It is therefore important to ensure that the maximum
operating frequency for a given MIC remains below the frequency associated with the
generation of this resonance, for the widest strip in the MIC. This resonance will be manifested
as a half-wave with its node in the centre and antinodes beyond the edges of the microstrip
(due to side-fringing) as shown in Figure 3.15.
The side-fringing equivalent distance is denoted by d (= 0.2h for most microstrips), and
the equation dictating the resonance is therefore:

λ CT
= w + 2d
2
= w + 0.4h (3.107(a))

x
w (w + 2d)

εr

Ey

x
λCT /2
0 x = w + 2d

Figure 3.15 Transverse microstrip resonance; standing voltage wave and


equivalent transverse electrical ‘line’
Signal Transmission, Network Methods and Impedance Matching 137

or, equivalently:

c
fCT = (3.107(b))
ε r (2w + 0.8h)

Slots in the microstrip, usually down the centre to cut into the longitudinal current, can
suppress this transverse resonance.

Self-assessment Problem
3.28 For the two microstrips in the previous problems (including the 80 ohm one);
calculate the frequencies of onset for the lowest order TM mode and for trans-
verse resonance. Compare these results and comment on the practical operating
implications.

3.6.4 Loss Mechanisms


In common with all types of electrical circuit elements, microstrips dissipate power. The
following three loss mechanisms are the most important:

1. Conductor
2. Dielectric
3. Radiation.

The first two, conductor and dielectric losses, are more or less continuous along the direction
of propagation of the electromagnetic wave in the microstrip transmission line. Radiation loss,
on the other hand, tends to occur from discrete apertures such as nominally open conductor
ends. There are also parasitic losses due to surface wave propagation, when this occurs.
Both conductor and dielectric losses are described in terms of contributions to the loss
coefficient α of the microstrip transmission line. Conductor loss is influenced strongly by
the skin effect and hence also by surface roughness (on the underside of the strip). Denoted
by αc, it is given to a good approximation by:

f  2   ∆  2 
α c′ = 0.072 λg 1 + tan −1 1.4    (3.108)
wZ0  π   δ s  

Self-assessment Problem
3.29 What approaches to design tend to lead to low α c?

The dielectric loss is a consequence of the dissipation mechanism within the dielectric of
the substrate, usually characterised by the loss tangent, tan δ. This loss is reduced by the fact
138 Microwave Devices, Circuits and Subsystems for Communications Engineering

that some of the energy is transported in air. The resulting loss, in decibels per microstrip
wavelength, is:

ε r (ε eff − 1) tan δ
α d′ = 27.3 [dB/microstrip wavelength] (3.109)
ε eff (ε r − 1)
Clearly, the only practical way to maintain low dielectric loss is to ensure that substrate
materials have low tan δ.
Typically microstrip structures have physically open sections and other discontinuities
(see Section 3.6.5). In general energy is radiated by such structures – although this energy
will be associated with near field components only, since the conducting top and side-walls
of the metallic box normally used to enclose the subsystem reflect any radiation and set up
standing waves or multiple reflections. Little of this energy thus escapes from the enclosed
module, but it does contribute to overall losses and should be minimised.
An open-ended microstrip can be treated as having an associated shunt admittance Y
given by:

Y = Gr + Gs + jB (3.110)

where Gr is the radiation conductance (modelling the radiation loss), Gs is the surface wave
conductance (modelling the surface wave loss) and B is the shunt susceptance. The radiation
conductance may be approximated by:

4πhweff
Gr Z0 ≈ (3.111)
3λ 20 ε eff

where weff is the effective microstrip width, defined by:


2
w eff = c2/4 f 2p εeff (3.112)

In this expression fp is the quantity defined by Getsinger, see Section 3.6.2. (This is slightly
non-linearly frequency dependent, i.e. dispersive.)

Self-assessment Problem
3.30 Assuming the same 31.3 ohm microstrip as defined for previous SAPs:

1. Calculate the conductor and dielectric losses at a frequency of 14 GHz for


a substrate with tan δ = 0.002. Compare your results and comment on the
practical implications. (Ignore surface roughness.)
2. Again, at a frequency of 14 GHz, calculate the equivalent radiation loss
resistance and sketch the complete equivalent circuit for a length of line with
one open end.

What would be the result of encouraging (instead of minimising) these radiation


losses? Is this approach at all useful at relatively low frequencies? Explain.
Signal Transmission, Network Methods and Impedance Matching 139

3.6.5 Discontinuity Models


Continuous, uninterrupted, sections of microwave transmission lines are unrealistic in practice
and networks must be created that involve graded or sudden changes in structure. These
changes, called discontinuities, are very important in MIC (and MMIC) design. The following
represent examples of microstrip discontinuities:

1. The foreshortened open end.


2. The series gap.
3. Vias (i.e. short circuits) through to the ground plane.
4. The right-angled corner or bend (unmitred or mitred).
5. The step change in width.
6. The transverse slit.
7. The T-junction.
8. Cross-junctions.

In order to show instances of several of these discontinuities a typical single-stage MIC


transistor amplifier (GaAsFET) layout is shown in Figure 3.16.
Modelling and computation of the effects of all these discontinuities have been the subject
of much research effort over the past four decades.
We shall restrict our considerations here to the following types of discontinuities:

1. The foreshortened open end.


2. Vias.
3. The mitred right-angled bend.
4. The T-junction.

3.6.5.1 The foreshortened open end


As already discussed, radiation and surface waves are generally launched from this type
of discontinuity. There is also, however, susceptance (B in the admittance expression) at

Input matching circuit


5
4
6
4
7

Output matching circuit


5
4
4
6
4
4
7

(1)
(5) (5) (4)

GaAsFET on
“Chip on Disc”
MIC mount
25 mil Alumina
substrate
(0.635 mm)

Figure 3.16 Single-circuit layout of a single-stage GaAsFET MIC amplifier


140 Microwave Devices, Circuits and Subsystems for Communications Engineering

SIDE ELEVATION

g
GAP

l
OPE
N
END

FRINGING E FIELD

(a) Physical open circuits

l
C2

Cf C1 C1

OPEN-CIRCUIT GAP
END CAPACITANCE CAPACITANCES
(b) Equivalent networks

Figure 3.17 Microstrip open end and series gap: (a) physical circuits, (b) equivalent networks

this plane, due mainly to capacitance resulting from the local electric field fringing from the
open end of the microstrip down to the ground plane. A practical example of this feature
(and also a series gap) is shown in Figure 3.17(a) and the equivalent lumped capacitance
associated with these structures is shown in Figure 3.17(b).
In some CAE software approaches the end-fringing capacitance, Cf, may be used directly
within the design. However, it is important to observe that this particular discontinuity
frequently occurs at the ‘end’ of an open circuit stub – which could be part of a filter or
a matching network (see Section 3.9 and subsequent sub-sections). In this case it is the
complete effective electrical length of the stub that is required in the design. This poses an
initial problem, because we have mixed types of circuit elements here – one distributed
(the microstrip itself ) and the other lumped (Cf ), Figure 3.18.
In terms of Cf directly it can be shown that this additional section of line, to be added in
series beyond the microstrip open end, is given by:

cZ0Cf
le 0 ≈ (3.113)
ε eff

Equation (3.113) is adequate providing the designer knows the accurate value of Cf for all
microstrips in the network. This is not particularly convenient, however, and an empirical
expression (that embodies only microstrip parameters) such as:

 ε + 0.3   w / h + 0.262 
le 0 = 0.412h  eff   (3.114)
 ε eff − 0.258   w / h + 0.813 
Signal Transmission, Network Methods and Impedance Matching 141

(a) Physical microstrip

(b) Transmission line


zo Cf with equivalent
end fringing
capacitance Cf

l eo (c) Transmission line


with equivalent
extra transmission
zo zo line of length l eo

Figure 3.18 Equivalent elements to represent the microstrip open end fringing field

is useful. Error bounds of approximately 5% apply when using this expression over a
wide range of types of microstrips on differing substrates. This degree of error should be
acceptable in most cases.

3.6.5.2 Microstrip vias


The provision of short-circuiting holes between conductor patterns is a familiar one applying
to circuits operating from DC to millimetre waves. Lower frequency circuits, including the
important multi-chip modules (MCMs) require such vias. They are also highly significant in
monolithic integrated circuits of all types – ICs and MMICs.
It is instructive to pose the question when is a short circuit not actually a short circuit?
The answer is, when the frequency is sufficiently high.
Even for a conducting hole prepared in an extremely thin substrate, the term ‘short circuit’
is only ever near-perfect at DC. As the frequency increases so losses due to the skin effect
mount and the inductive reactance becomes increasingly significant.
A short-circuiting conductive hole, manufactured at the otherwise ‘open’ end of a microstrip
is shown in Figure 3.19.
A useful optimising condition that has been derived for this structure is:
2
 w   πd 
ln  eff  ≈  e  (3.115)
 π de   weff 

where de = 0.03 + 0.44d, d is the actual (physical) hole diameter and weff is the effective
microstrip width. Provided this condition is satisfied, the hole should remain an acceptable
broadband short circuit over the frequency range 4 to 18 GHz. Calculations using Equation
142 Microwave Devices, Circuits and Subsystems for Communications Engineering

Microstrip

Ground
plane

Short-circuiting
metallized hole

Figure 3.19 A shunt metallised hole in microstrip (at an otherwise ‘open’ end)

(3.115) require substitution and iteration, since the equation is transcendental in both de
(and hence d) and weff .

3.6.5.3 Mitred bends


If a sudden bend is created in microstrip, sharp corners are inevitable on both the inside
and the outside of the bend. These give rise to substantial current discontinuities because the
major proportion of the total current flows in the outside edges. In turn, these current dis-
turbances lead to excess inductances. Another, equally valid, way of viewing the situation is
to appreciate that such sharp bends lead to significant mismatches, even when both ends of
the complete microstrip line are terminated. It is evident, therefore, that an attempt must be
made to reduce the effect of the bend discontinuity. The most important approach to this
problem is to chamfer (or mitre) the ‘long’ outside edge, as shown in Figure 3.20.
On first consideration it might be thought that increasing advantage would be obtained the
greater the degree of chamfer. However, in the limit this results in local microstrip-narrowing
and current-crowding in the cornered region – so there is actually an optimum degree of
chamfer. It has been shown that this optimum is, to a good approximation, given by:

b ~ 0.57w (3.116)

3.6.5.4 The microstrip T-junction


In many instances it is necessary to branch from one microstrip into another. Examples
include stubs and branched signal routing. Such branching is often achieved using a right-
angled (T-shaped) junction, Figure 3.21. These junctions involve inherent discontinuities
that need to be modelled.
In common with the microstrip bend, and for identical reasons, we must include series
inductance to account for current disturbances, and electric field distortion means that
capacitance has to be introduced locally at the junction plane. It is also necessary to account
for the effects of general impedance loads introduced at some plane along the branching
microstrip line. This is accommodated by means of an equivalent transformer element
that transforms this impedance so that its effect is equivalently introduced into the main
microstrip line.
Signal Transmission, Network Methods and Impedance Matching 143

lc lc
2 2
w

zo jB zo
b

w p′ p p
(b) Equivalent circuit
p
(a) Structure and nomenclature

B lc
0.3 Y0 h

B
Y0 0.2
and
lc 0.1
p
0
0 0.2 0.4 0.6 0.8
chamfer fraction (1 – b )
2w
(c) Parameter trends

Figure 3.20 The chamfered (or mitred), right-angled, microstrip bend and relationships

p1 L1 L1 p1
p1
From To ‘main’
generator load C
w1 w1

w2
p2 1

L n
Ideal
Secondary load transformer
p2

(a) Structure and


nomenciature (b) Equivalent circuit
p2

Figure 3.21 The microstrip T-junction and its elementary equivalent circuit

There have been some difficulties in using this equivalent circuit, especially in terms
of accurately modelling the transformer (which requires a frequency-dependent turns ratio).
Driven by these problems, research has yielded semi-empirical design expressions based
upon including effective shifts in reference planes in this structure. The nomenclature for
the reference planes is shown in Figure 3.22.
144 Microwave Devices, Circuits and Subsystems for Communications Engineering

d1
w1

Z0(1) Z0(1)
d2 d2
w2 Z0(2)

Figure 3.22 Reference planes and impedances for the T-junction

In all the following expressions, relating to the T-junction, the subscripts 1 and 2 denote
microstrip number 1 (the main line) and microstrip 2 (the branch line), the impedances are
characteristic impedances for the lines, and the capacitance C is the local equivalent junction
capacitance taking account of the electric field disturbances. In this model effective microstrip
widths (defined earlier) are also used, and their values for either line are given by:


weff 1,2 = (3.117)
Z0(1,2 ) ε eff 1,2

where η is the impedance of free space (376.7 Ω). The turns ratio, n, reference plane dis-
placements and capacitance are given by the following (complicated) algebraic relationships:
2
 sin{π (weff 1 / λg1 )( Z0(1) / Z0(2 ) )} 
n2 =   [1 − {π (weff 1 / λg1 )(d2 / weff 1 )} ]
2
(3.118)
 π (weff 1 / λg1 )( Z0(1) / Z0(2 ) ) 
The displacement of the reference plane for the primary line (1) is:

d1 Z
= 0.05 0(1) n2 (3.119)
weff 2 Z0(2 )

and the displacement of the reference plane for the secondary arm (2) is:

d2   2weff 1 
2
 Z0(1)   Z0(1)   Z0(1)
= 0.5 −  0 .076 + 0 .2  λ  + 0.663 exp  −1.71 Z  − 0.172 ln  Z   Z (3.120)
weff 1   g1   0( 2 )   0( 2 )   0( 2 )

The shunt capacitance is determined by the following expression for the condition Z0(1) /Z0(2)
≤ 0.5:

ω Cλg1  2w  Z
=  eff 1 − 1 0(1) (3.121)
Y0(1) weff 1  λg1  Z0(2 )

and, for the condition Z0(1) /Z0(2) ≥ 0.5, by:

ω Cλg1  2w  Z 
=  eff 1 − 1  2 − 3 0(1)  (3.122)
Y0(1) weff 1  λg1  Z0(2 ) 
Signal Transmission, Network Methods and Impedance Matching 145

Discrepancies resulting from the inherent approximations in these expressions increase when,
for the main line, twice the effective width divided by the wavelength exceeds about 0.3.
It has been found by measurements that the d2 prediction by Equation (3.120) is typically
too large at microwave frequencies – by as much as a factor of two or more. (At microwave
frequencies, therefore, halving d2 generally provides a better result.)

Self-assessment Problem
3.31 For the same microstrip line specified for earlier questions (the 31.3 ohm line),
calculate:
(a) The equivalent open end effective length. What is the electrical length, in
degrees, of this at a frequency of 20 GHz? How does this electrical length
compare with that of a one-eighth wavelength of the microstrip line itself ?
(b) The physical diameter of an optimum broadband short-circuiting via.
(c) The width of microstrip conductor remaining after chamfering a right-angled
bend. (Also calculate this width for a 90 ohm microstrip – other parameters
the same – and comment on any obvious practical problem arising.)

3.6.6 Introduction to Filter Construction Using Microstrip


Compared with coaxial lines, waveguides, or dielectric resonators, most planar structures,
including microstrip, exhibit relatively low Q-factors (over 100 is often difficult to realise).
This is a serious limitation in respect of several types of filter design. With specifications
suitable for many purposes, however, a range of classic filters can be constructed in microstrip.
(Where Q-factors much above 100 are absolutely essential, then alternative technologies
must be selected, e.g. dielectric resonator ‘pill’ chips surface mounted on to otherwise
microstrip MICs.)
We shall exclusively discuss low-pass filters at this point, because the design require-
ments for these encompass the microstrip techniques described previously.

3.6.6.1 Microstrip low-pass filters


The approach here begins by taking the model of lumped component low-pass filter (LPF)
design comprising a cascade of C-L-C sections. The aim is to first determine the (often
unrealisable) set of lumped component values, e.g. 1 pF, 3 nH, 2 pF, and then to transform
these values into realisable microstrip sections. The general lumped network topology is
shown in Figure 3.23(a) and the derived distributed (microstrip) configuration is shown in
Figure 3.23(b).
The element values in the lumped network are obtained by conventional filter synthesis.
This starts with low-pass filter synthesis determined by the required insertion loss char-
acteristic and then proceeds to the microwave values by performing suitable frequency
transformations (e.g. Butterworth, Chebychev). We assume here that the calculations have
been performed and that the microwave-version lumped LPF topology is known (for a single
146 Microwave Devices, Circuits and Subsystems for Communications Engineering

L2 L4

C1 C3 C5

(a)

l1 l3 l5

l2 l4
50 Ω WL WL 50 Ω
WC WC WC

L2 L4

C1 C3 C5
(b)

Figure 3.23 Lumped network topology (a) and microstrip configuration (b) for a low-pass filter

Z0 sinhγ l
=
z

γ γ
Z0 coth l = z z = Z0 coth l
2 2

Figure 3.24 Equivalent π-network of impedance elements representing a length l of


transmission line having propagation coefficient γ

or double-terminated filter, as desired). The next stage is to realise each inductive and
capacitive element in microstrip form.
Fundamentally, we have a situation where inductive and capacitive lines are adjacent,
which can be represented by the equivalent π-network shown in Figure 3.24.
In this approach to LPF design the effects of losses are neglected so that all elements are
(as required by the lumped element model) reactive/susceptive. The basic lumped element
π-network, its lumped-distributed equivalent and the final microstrip form of the structure
are shown in Figure 3.25.
Fundamental transmission line theory shows that the input reactance of a line of length l
is given by:

 2π l 
XL = Z0 sin   (3.123)
 λg 
Signal Transmission, Network Methods and Impedance Matching 147

l
( jx) (a) Lumped
C C
circuit

(b) Lumped-distributed
C Z0 C
equivalent

l
(c) Microstrip form
High
Low impedance impedance Low impedance

Figure 3.25 Inductive line with adjacent capacitive lines

which can be re-arranged to yield the length of the required section of line:

λg  ω L
l= sin −1   (3.124)
2π  Z0 

When (and only when) the section of line is electrically short (much less than a quarter
wavelength) it is possible to simplify Equation (3.124):

fλg L
l≈ (3.125)
Z0
In most cases, however, the full expression given in Equation (3.124) will be required.
The shunt end susceptances are given by:

1  πl 
BL = tan   (3.126)
Z0  λg 
which for short electrical lengths yields the following approximate expression for capacitance:

1
CL ≈ (3.127)
2 f Z0 λg
In addition to these components, when realising the physical microstrip line there will
also be the discontinuity elements due to the step in impedances. We did not consider this
particular discontinuity, but where the step is large the main capacitive element of the dis-
continuity is approximately that of the wide adjacent capacitive microstrips.
A short length of microstrip line having relatively low characteristic impedance (i.e. a
wide line) is predominantly capacitive and its shunt susceptance is given by:

1  2π l 
B= sin   (3.128)
Z0  λg 
148 Microwave Devices, Circuits and Subsystems for Communications Engineering

which can be re-arranged to yield the length of this short line:

λg
l= sin −1 (ω CZ0 ) (3.129)

Again, since these lines are always electrically short the following approximation holds with
reasonable accuracy:
l ≈ f λ g Z0 C (3.130)
The inductances associated with this predominantly capacitive line are usually negligible.

3.6.6.2 Example of low-pass filter design


We start with the basic specification, which is a 3 dB cut-off frequency of 2 GHz. (Note:
cut-off frequencies are not always specified at the 3 dB level for filters – always check.)
The design is based on a 5th order Butterworth response, and a double-terminated 50 ohm
system.
We are assuming that the fundamental calculations have been completed as far as the
microwave prototype lumped network. Following the nomenclature of Figure 3.23(a) the
values are:
C1 = 0.98 pF; L2 = 6.4 nH; C3 = 3.18 pF; L4 = 6.4 nH; C5 = 0.98 pF
(and, to repeat, a 50 ohm system extends on both the signal input and output ends of this
circuit).
All the design expression given above are now employed to develop the required microstrip
line lengths, on an alumina substrate of relative permittivity 9.6 and thickness 0.635 mm.
The final 2 GHz LPF layout is shown in Figure 3.26.
In practice this type of design is generally developed using proprietary RF/microwave
CAE software.
There are many further possibilities with LPF (and other) filter designs. Notably, the low-
impedance line sections can be replaced with ‘quasi-lumped’ capacitive pads – usually circular
configurations. It is important to appreciate that these are indeed strictly ‘quasi-lumped’
because, as the frequency is increased, resonant modes can occur within the pad. Obviously
this complicates the design.

3.7 Coupled Microstrip Lines


When any forms of transmission line, specifically TEM propagating (or closely related)
forms, are located adjacently, then a fraction of the signal energy travelling on one line
is coupled across to the other line. This is edge-coupling or parallel coupling and accounts
for the unwanted cross-talk experienced in many circuits. The coupling process results in
contra-directional travelling waves; the coupled wave travels in the opposite direction to
that of the incident wave.
Such coupling is important in many instances, extending from directional couplers that
are required in many applications, to parallel coupled resonators that are highly significant
in parallel-coupled bandpass filters.
Signal Transmission, Network Methods and Impedance Matching 149

16.705 mm 16.705 mm

9.922 mm

0.632 mm 0.0812 mm 0.0812 mm 0.632 mm

6.19 mm 6.19 mm

0.935 mm 2.850 mm 0.935 mm

Figure 3.26 Layout of the 2 GHz Butterworth microstrip LPF

There is considerable flexibility in the configuration of this type of coupling – for instance,
the lines do not have to be identical, they could have different geometries.
Here we shall concentrate upon the pair of identical parallel-coupled microstrips. The
configuration is shown in Figure 3.27.
Throughout the following subsections it is important to remember that electromagnetic
transmission on microstrips is quasi-TEM and not pure TEM. This has far-reaching con-
sequences for the behaviour and achievable specifications of microstrip coupled lines. Analysis
of coupled microstrips begins with the concept of superimposing the effects of two separately
identifiable modes associated with this situation – the so-called even and odd modes. This
concept is vital to understanding, and designing, any component using coupled microstrips.

Microstrip lines

Substrate

Ground plane

Figure 3.27 Cross-section showing a pair of parallel-coupled (or ‘edge-coupled’) microstrips


150 Microwave Devices, Circuits and Subsystems for Communications Engineering

E-field
H-field

+ + + −

(a) Even mode (b) Odd mode

Figure 3.28 Even- and odd-mode field distributions for parallel-coupled microstrips

3.7.1 Theory Using Even and Odd Modes


The concept of these modes arises from considering the extremes of polarisation for the
driving and coupled voltage on each microstrip. The aim is to eventually use the results from
considering these extremes to develop design expressions for the coupled situation (such as
degree of coupling, directivity, etc.).
The even mode is defined as the field situation produced when the voltage of each microstrip
has identical polarity (either both positive or both negative). In contrast, the odd mode is
defined as the field situation that exists when the voltages on each microstrip are of opposite
polarity. These two extreme situations are shown in Figure 3.28 (in which only outlines of
the electric and magnetic fields are considered).
Consequently, instead of a single characteristic impedance, phase velocity, capacitance,
effective microstrip permittivity and electrical length (θ ) being associated with a single
microstrip as covered in some detail in Section 3.6 and subsequent subsections, we have two
definitions for each of these quantities – one for each mode. The notation therefore becomes:

Z0e and Z0o for characteristic impedances


Cie and Cio for capacitances
εeffe and εeffo for effective relative permittivities

Extensive analysis has been undertaken in order to generate design data for these parameters
under a wide variety of microstrip conditions. We shall not consider this analysis in detail
here, but we shall need to be clearly aware of the results, consequences and limitations.
First let us consider the overall nomenclature associated with these coupled microstrips.
This differs only by virtue of having one extra physical parameter – namely the strip
separation, s. Otherwise the microstrips have widths, w, and are manufactured on a substrate
of relative permittivity εr and thickness h as for single microstrip. The cross-section defining
physical dimensions is shown in Figure 3.29.
Note that there is one further important dimensional parameter not shown in Figure 3.29,
and this is the length of the coupled region, usually denoted as l.
All the parameters are inherently dependent on the degree of coupling required. This
might seem obvious but it is important to appreciate that this fact extends to the widths of
the microstrips as well as the separation s and the length l. Also, as with single microstrip, all
the parameters of the coupled microstrips are strictly frequency-dependent and so dispersion
exists in this structure – but with important differences compared to the single microstrip.
Signal Transmission, Network Methods and Impedance Matching 151

h S
W W

Figure 3.29 Nomenclature for the cross-sectional dimensions of parallel-coupled microstrips

Modelled on the results for single microstrip, the odd- and even-mode characteristic
impedances are given by:

1
Z0 o = (3.131)
cC1o ε effo

and:

1
Z0e = (3.132)
cC1e ε effe

To a limited extent, with some difficulties in terms of design use due to curve convergence,
interpolation, and other approximations, the graphs of Figure 3.30 can be used for approximate
designs. They are also useful for general guidance.

400 200
s
350
h
300 150 0.05
0.2
250 0.5
1.0
Characteristic impedance (ohms)

Characteristic impedance (ohms)

200 100 2.0



90 2.0
80 1.0
150 s
70 0.5
Odd mode even mode

h
0.05 60 0.2
0.2 s
0.5
100 50 h
Odd mode even mode

1.0
2.0 0.05
90 ∞ 0.2
80 2.0 40 0.05 0.5
1.0
1.0 2.0
70 0.5 ∞
2.0
60 30 1.0
0.2
0.5
50 εr = 1.0 εr = 9.0 0.2
(i.e. for air or free-space) 0.05
40 0.05
0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
w/h w/h
(a) (b)

Figure 3.30 Families of curves for even- and odd-mode characteristic impedances for coupled
microstrips in (a) air and (b) on a substrate having relative permittivity 9.0
152 Microwave Devices, Circuits and Subsystems for Communications Engineering

We now consider some design equations applicable to a microstrip coupler and also to the
use of coupled microstrips for the realisation of components such as band-pass filters.
If we are aiming to design a microstrip coupler based on this configuration, then we start
with the required coupling factor. We shall use the symbol C for this parameter as a linear
ratio, and the symbol C′ when it is specified in dB (the usual practice). All the following
simple and direct relationships are for mid-band.
The coupling factor is the given by:

Z0e − Z0 o
C ′ = 20 log10 [dB] (3.133)
Z0e + Z0 o

There also exists an approximate impedance interrelationship that is required to develop


expressions for the even- and odd-mode characteristic impedances and which may be used
as a check on relative values during a design process. This is:

Z 20 ≈ Z0e Z0o (3.134)

The error in this expression increases strongly as the degree of coupling increases.
Using Equations (3.133) and (3.134) the important even- and odd-mode characteristic
impedances can readily be derived. These are:

1 + 10 C′/20
Z0e ≈ Z0 (3.135)
1 − 10 C′/20

1 − 10 C′/20
Z0 o ≈ Z0 (3.136)
1 + 10 C′/20

Equations (3.135) and (3.136), are the essential starting expressions for proceeding with
a microstrip coupler design. Almost always the designer will know the single microstrip
characteristic impedance Z0 and also the required mid-band coupling factor C′ (the expressions
as written here automatically convert the dB value of C′ into its linear equivalent).
A completely accurate and universal relationship for Z0 as a function of Z0o and Z0e is
obtainable from fundamental analysis of the coupled microstrip system. This starts with the
two-port ABCD matrix for the voltages and currents on each line, incident (denoted by
subscript 1) and coupled (denoted by subscript 2):

V1   cos θ e + cos θ o j( Z0e sin θ e + Z0 o sin θ o )  V2 


 =   (3.137)
 I1   j(Y0e sin θ e + Y0 o sin θ o ) cos θ e + cos θ o   I2 

(See Section 3.8.1 for a revision of matrix representations for two-port networks.) For a
theoretically perfect match (also infinite isolation) the diagonal terms in the admittance
matrix must be forced into equality. We need this condition and when invoked it leads to the
following expressions:

Z0e Z Z Z
sin θ e + 0 o sin θ o = 0 sin θ e + 0 sin θ o (3.138)
Z0 Z0 Z0e Z0 o
Signal Transmission, Network Methods and Impedance Matching 153

or:

Z0e sin θ e + Z0 o sin θ o


Z 02 = Z0e Z0 o (3.139)
Z0e sin θ o + Z0 o sin θ e

There are two highly significant features in these expressions, both important to understanding
couplers (and coupled microstrips generally). These are:

1. Because of the differing field configurations in the even and odd modes, the electrical
lengths θe and θo can never be exactly equal, including at mid-band.
2. These electrical lengths (θe and θo) are naturally (differing) functions of frequency.

These features tell us that mid-band design will never be precise and will require com-
pensation in order to adjust the coupling factor and increase the directivity. They also tell us
that we are dealing with a frequency-dependent device and therefore some broad-banding
approach is desirable.
Returning to the mid-band design approach, for moderate to relatively loose couplings
(typically 10 dB and looser), we can use expressions for the characteristic impedances as
functions of the speed of light c and the capacitances of the individual microstrips both in
air and with the substrate dielectric – for each mode. The effective microstrip permittivities
are also the ratio of these capacitances – again for each mode:

1
Z0e = (3.140)
c CeCe1

and:

1
Z0 o = (3.141)
c CoCo1

Also:

Ce
ε effe = (3.142)
Ce1

and:

Co
ε effo = (3.143)
Co1

There is a powerful approximate synthesis approach for coupled microstrips, which is useful
in both coupler and band-pass filter design. This is presented next.
The aim here is to obtain first-cut approximate results for w and s. (More accurate figures
can then subsequently be derived by analysis, re-calculation of the characteristic impedances,
and iteration – straightforward on a computer.)
154 Microwave Devices, Circuits and Subsystems for Communications Engineering

There are two main stages in this approximate synthesis:

1. Determine (intermediate step) shape ratios for equivalent single-modelled microstrips:


(w/s)se and (w/s)so. All subscripts ‘s’ here refer to this single microstrip model.
2. Use these shape ratios to find the final coupled microstrips widths w and separation s.

For stage 1 it is necessary to use halved values of the characteristic impedances. For the
single microstrip ratio (w/s)se therefore we have:

Z0e
Z0 se = (3.144)
2

and, for the single microstrip shape ratio (w/s)so we must use:

Z0 o
Z0 so = (3.145)
2

Explicit equations are available for the single microstrip shape ratios and also for s/h:

 w 2  2 d − g + 1
= cosh −1   (3.146)
 h  se π  g +1 

If εr ≤ 6:

 w 2  2 d − g − 1 4  w/h 
= cosh −1   + cosh −1 1 + 2  (3.147(a))
 h  so π  g −1   ε   s/ h 
π 1+ r
 2

or if εr ≥ 6:

 w 2  2 d − g − 1 1 −1 
w/h 
= cosh −1   + cosh 1 + 2  (3.147(b))
 h  so π  g −1  π s/ h 

where:

 πs 
g = cosh (3.148)
 2h 

and:

 w πs 
d = cosh π + (3.149)
 h 2h 

In order to use these expressions as a design aid, Equation (3.146) must be solved simultane-
ously with either Equation (3.147(a)) or (3.147(b)) – depending on the permittivity of the
substrate used. However, the second terms in Equations (3.147(a)) and (3.147(b)) may often
Signal Transmission, Network Methods and Impedance Matching 155

be ignored (especially with tight coupling, since then w/h will generally be some factor
greater than s/h). Under this condition a direct formula for s/h is obtained:

s 2  cosh{(π / 2)(w / h)se} + cosh{(π / 2)(w / h)so} − 2 


= cosh −1   (3.150)
h π  cosh{(π / 2)(w / h)so} − cosh{(π / 2)(w / h)se} 

Although errors exceeding 10% occur when this technique is used alone, it is possible to
obtain much improved accuracy by starting with these approximate relations and then solving
accurate analysis expressions, followed by iteration. This approach is summarised below:

1. Starting with the desired Z0e and Z0o (from C′ or from filter design), find initial values
(w/h)1 and (s/h)1 using the approximate synthesis approach.
2. Apply these first (w/h)1 and (s/h)1 quantities to re-calculate Z0e and Z0o using relatively
accurate analysis formulas.
3. Compare the two sets of Z0e and Z0o values, noting the differentials, and iterate the above
calculations until the two sets of values are within a specified error limit. (In the case of
a coupler, also compare values of the coupling factor C required.)

The values of w and s at this final stage are the closest to those required in the design.
The above expressions can be combined and data plotted in the form of the graph shown
in Figure 3.31.

5.0 e Odd mode (o)


5
10 6 Even mode (e)
7
4.0
e
w Quantities all reter
5
h coupled to w/h single
o 86
7 obtained from a
3.0 single microstrip
e
5 line design
o 66 procedure
7
2.0 o
e

e o 4
1.0 o 3
e
e o 2
o 1
0
0 0.5 1.0 1.5 2.0
s
h

Figure 3.31 Family of curves for use in the approximate synthesis design
(strictly applicable only for εr = 6)
156 Microwave Devices, Circuits and Subsystems for Communications Engineering

This family of curves can often be particularly difficult to use because values are frequently
within crowded regions of the graph.

Self-assessment Problem
3.32 A microstrip coupler, using parallel-coupled lines, is required to have a coupling
factor of 10 dB. (Note that this means the voltage level coupled into the coupled
line is −10 dB (i.e. 10 dB down) with respect to that on the incident line.)
The single microstrip feeder connecting lines have terminated characteristic
impedances of 50 ohms. The substrate has a relative permittivity of 9.0 and a
thickness of 1 mm. Calculate the width and separation of the coupled lines using:
(a) The curves shown in Figure 3.30, and (b) the approximate synthesis curves
of Figure 3.31.

3.7.1.1 Determination of coupled region physical length


So far, in the design approaches for parallel-coupled microstrips, we have only concentrated
on the determination of cross-sectional dimensions: w and h. A further quantity must be
established to complete designs, namely the physical length of the coupled region, l.
In mid-band, i.e. for the maximum coupling condition, the electrical length of the coupled
region should be an integral number of quarter wavelengths:

(n − 1)λ g /4 (3.151)

To avoid harmonic effects (repeated responses at higher frequencies) it is best if the region
is simply one-half wavelength, i.e. n = 2, and the length is simply λ g /4. However, at high
frequencies this leads often to physically short structures that may be difficult to manu-
facture and n = 3, 4, etc. may be desired. It is pertinent to ask the question, here, what is the
value λ g? Remember, there are the two modes (even and odd) and they have different phase
velocities and hence different wavelengths.
Various methods for dealing with this problem have been put forward. We shall describe
the weighted mean approach here, in which the characteristic impedances are also invoked.
Using this approach the weighted mean phase velocity is:

Z0e + Z0 o
vn = (3.152)
( Z0e / ve ) + ( Z0 o / vo )

and (following weighting) the appropriate microstrip wavelength is given by:

vn
λgn = (3.153)
f0

This value is then used in Equation (3.151) to find the appropriate length. Some discontinuity
will exist around the end regions of the coupler, where the single microstrips connect. This
may be minimised by chamfering the entering corners (see Section 2.5.3).
Signal Transmission, Network Methods and Impedance Matching 157

Self-assessment Problem
3.34 Calculate the quarter-wave coupling region length for the coupler of the previous
SAP, operating at a centre frequency of 5 GHz.

3.7.1.2 Frequency response of the coupled region


As remarked earlier in this section all the parameters of the coupled region are in fact
frequency dependent due to two underlying phenomena:

1. The inherent frequency behaviour of any TEM parallel coupler.


2. The dispersion in the quasi-TEM microstrips differing between even and odd modes.

The first contribution to the frequency dependence, the TEM situation, will actually be inter-
preted to encompass this quasi-TEM structure and the relevant expression obtained from
fundamental analysis is:

jC sin θ
C(θ ) = C( jω ) = (3.154)
1 − C cos θ + j sin θ
2

In this expression C is the mid-band value of the coupling factor already described in some
detail. The phase angle θ is the general, frequency-dependent, quantity θ = βl = 2πl/λg.
While l will obviously be constant, β will vary since the wavelength λ g will change with
frequency.
For the ideal matched (i.e. terminated) coupler of this type the frequency response,
neglecting dispersion, will closely follow that shown in Figure 3.32.
Unlike single microstrips, for coupled microstrip lines dispersion affects the even and odd
modes in different ways. Referring back to Figure 3.28 it can seen that the field distributions

Z0e – Z0o
|C(jω)| C= (max)
Z0e – Z0o

ω
mid-band

Figure 3.32 Frequency response for a loosely-coupled ideal matched coupler


158 Microwave Devices, Circuits and Subsystems for Communications Engineering

are markedly different for each of these modes, including at low frequencies. It should
therefore be clear that, as frequency increases and the coupling beneath the strips intensifies
(see single microstrip dispersion in Section 3.6.2), the detailed changes in field distributions
must differ between the even and odd modes.
As well as functions of frequency and substrate thickness, the effective microstrip per-
mittivities for the even and odd modes are now also functions of the low frequency limiting
values and the characteristic impedances (obtained by analysis).
Getsinger has shown how his (single microstrip) dispersion expressions may be adapted
to approximately predict this dispersive behaviour. The key to this follows from studying the
effects on characteristic impedance substitution, summarised thus:

1. For the even mode the two microstrips are at the same potential, but the total current
flowing down the system is twice that of a single strip with otherwise identical structure
(twice because of the paralleled conductors). This implies that the correctly substituted
impedance should be one half of Z0e. For the even mode case substitute, for Z0, 0.5 Z0e.
2. For the odd mode the two microstrips are at opposite potentials and the potential dif-
ference between the strips is therefore twice that of a single microstrip. However, the
current is the same as for a single microstrip, and the correctly substituted impedance
should therefore be twice Z0o. For the odd mode case substitute, for Z0, 2Z0e.

These two substitutions should be made in the formulas for Getsinger’s single microstrip
line case (see Section 3.6.2).
This approach is still approximate, including at relatively low microwave frequencies, and
it should only be used with caution. More accurate expressions are available in the literature,
but these remain elaborate algebraic functions with extended numerical coefficients. They
are only really useful, therefore, for programming into CAE software routines, and not for
‘hand calculations’.

Self-assessment Problem
3.36 Considering again the coupler structure designed in the previous SAP, calculate
the length of the coupled region assuming the mid-band operating frequency to
be 12 GHz. Use Getsinger’s approach for dispersion. Note the differences between
the effective relative permittivities and compare the wavelength calculated here
with that calculated using the rough approximation: (5/12)λ g, where λ g is the
value for 5 GHz determined in the earlier SAP.

3.7.1.3 Coupler directivity


Ideally, a coupler should only couple signal energy from its transmission port (i.e. directly
through port 2) and its coupled port (port 3). However, in practice, ‘stray’ energy is also fed
to the fourth port.
Signal Transmission, Network Methods and Impedance Matching 159

The ability of a coupler to reject unwanted signal energy from being coupled back from
its fourth port is referred to as its directivity. Working in terms of voltages the definition of
directivity is:

V4
D= (3.155)
V3

The analysis leading to expressions for this directivity, derived from the fundamental equa-
tions governing the coupled line behaviour, is quite complicated. A more amenable set of
expressions is:
−2
 4 |ζ | 
D= 2 
(3.156)
 π∆ (1 − | ζ | ) 
in which:

 ρe   ρ o 
ζ =  −  (3.157)
 1 + ρ e2   1 + ρ o2 

where:
Z0e − Z0
ρe = (3.158(a))
Z0e + Z0

Z0 o − Z0
ρo = (3.158(b))
Z0 o + Z0

and:
λgo
∆= −1 (3.159)
λge
Calculations using these expressions reveal that the best values of directivity achievable
with the basic parallel-coupled structure are only around 12 or 14 dB, which is within the
same region of value as the coupling factor in many instances, making it unacceptable for
most applications. Techniques have therefore been developed for compensating for the
coupler in order to improve performance.

3.7.1.4 Coupler compensation by means of lumped capacitors


This is arguably the most sophisticated method, in the engineering sense, for compensating for
the behaviour of a microstrip coupled-line coupler. If lumped capacitors are introduced any-
where across the coupler they will only function in the odd mode, because only then is there
a potential difference across the capacitors. Furthermore, if such capacitors are implemented
across the end planes of the coupler they will not unduly disturb the main coupled region
effects. The configuration is shown, in network outline, in Figure 3.33. (There are actually other
advantageous performance features resulting from the introduction of such capacitors.)
160 Microwave Devices, Circuits and Subsystems for Communications Engineering

Coupled length
l

CL CL

Figure 3.33 Schematic diagram of coupled microstrips, with compensating capacitors


across the ends

The analysis of the circuit begins with expanding the ABCD matrix for this network (refer
to Section 3.8.1 for a review of ABCD parameters). Again, this analysis is fairly lengthy and
we shall only state the principal results here. The capacitance value is given by:

 π ε effo 
cos  
 2 ε effei 
CL = (3.160)
2 Z0 oiω

in which the final ‘i’ subscripts for the even-mode relative permittivity and the odd-mode
characteristic impedance refer to the ideal case (equal even- and odd-mode electrical lengths)
and ω = 2π f.
In practice, the capacitance values are usually much less than 1 pF (typically 0.05 pF or
so) and these can be difficult to realise except by precision thin-film interdigital approaches
– a technology that also leads to small gap dimensions.
For 10 dB couplers the directivity exceeds 20 dB over, typically, an 8 to 20 GHz band
(which is very broadband). Input VSWR maximises at around 1.4. The broadbanding appears
to be an essential unexpected advantage of this compensation method.
There are other ways in which these couplers can be compensated, notably using a dielectric
overlay which serves to almost equalise the phase velocities. This overlay may be introduced
as a dielectric layer co-fired during the manufacturing process (thin film or thick film), or it
could be introduced in the form of a surface-mounted chip. Some excellent results have been
achieved using this technique. A combined approach could also be explored in which both the
compensating lumped capacitors and the dielectric overlay might be introduced together.
Given the limitations of practically any variation on the parallel-coupled microstrip coupler,
especially where relatively tight coupling is required (e.g. 3 dB or greater), radically different
configurations are available. One of these (not strictly microstrip) is the combined directional
coupler. This uses a cascade of interdigital coupling elements (typically well over 10) along
an otherwise microstrip-like arrangement. This enables coupling factors approaching 0 dB to
be achieved and up to at least a three-octave bandwidth at low microwave frequencies. The
viability of this approach is limited to low microwave frequencies because of the essentially
quasi-lumped technology.
For fairly tight coupling, typically around 3 dB or tighter, over at least one octave of band-
width, the Lange coupler remains an excellent choice. This is covered in the next section,
which also deals with some so-called ‘hybrid’ couplers.
Signal Transmission, Network Methods and Impedance Matching 161

Bonding wires connecting


alternate fingers
Weld, ultra-sonic Isolated
Input 1 4
bond, etc.

Coupled 3 2 Direct

Figure 3.34 The basic form of the Lange coupler

3.7.2 Special Couplers: Lange Couplers, Hybrids and Branch-Line Directional Couplers
There have been many instances throughout the history of scientific and engineering inventions
where an essentially heuristic or empirical effort has led to exciting and sustained patents.
This is true of the Lange microstrip coupler, invented by Julius Lange while working at
Texas Instruments in 1961. The Lange coupler was first reported at a microwave conference
and initial results indicating its useful performance were shown – but for many years no formal
design equations were available and the realisation of these types of couplers remained
empirical. The general arrangement of the Lange coupler is shown in Figure 3.34.
It is immediately obvious that the Lange coupler structure differs greatly from the simple
parallel-coupled microstrip structure. The principal differences are:

1. There are interleaved, quarter-wave, coupling sections. (Two are shown here – in practice
there are many more.)
2. Bond wires straddle alternate coupling elements.

The first aspect (1) indicates that we are still principally interested in a quarter-wave coupling
effect (i.e. a quadrature hybrid) at mid-band. The second aspect is arguably the most innovat-
ive, namely, the introduction of bond wires, although this complicates the manufacturability
of the device.
The interdigital nature of the quarter-wave coupling sections ensures that substantial
power transfer can occur over a broad band, and the wires serve to transfer further power –
taking up what otherwise would be open ends of coupling microstrip fingers. It is essential
to ensure that the bond wires remain much less than eighth-wavelengths at the highest
design frequency, otherwise they will present serious design problems. In any case the
inductance of these bond wires must be minimised.
Basic design expressions, enabling the ‘standard’ parallel-coupled microstrip design already
covered to be used for the Lange coupler are:

Z0e Z0 o ( Z0e + Z0 o )2
Z 02 = (3.161)
{Z0e + (k − 1)Z0 o}{Z0 o + (k − 1)Z0e}
162 Microwave Devices, Circuits and Subsystems for Communications Engineering

This equation provides the relationship defining all the impedances. The voltage coupling
factor is given by:

C = 10−{voltage coupling ratio in dB/20}


(k − 1)Z 02e − (k − 1) Z 02o
= (3.162)
(k − 1)( Z 02e + Z 02o ) + 2 Z0e Z0 o

and the expressions for the even- and odd-mode characteristic impedances are:

C+q
Z0e = Z0 o (3.163)
(k − 1)(1 − C)

and:

1 − C (k − 1)(1 + q)
1/ 2
Z0 o = Z0 (3.164)
1 + C (C + q) + (k − 1)(1 − C)

where q is related to the number of fingers in the design (k) and the voltage coupling ration
(C) by:

q = {C2 + (1 − C2)(k − 1)2} (3.165)

The length of the entire coupler should be a quarter-wavelength at the lowest frequency in
the desired band, whereas the length of the intermediate fingers should be made a quarter-
wavelength at the highest frequency involved.
In the Lange coupler the even- and odd-mode phase velocities (and hence wavelengths,
etc.) are very nearly equal, as required in the ideal case. Double octave bandwidths, e.g. 10
to 40 GHz, are achievable.

Self-assessment Problem
3.37 (a) Suggest, with an explanation, a method for reducing the inductance of the
bond wires in Lange couplers.
(b) Determine the coupling conductor widths and separations for a Lange coupler
having the following characteristics:

1. Coupling factor = −10 dB.


2. Centre frequency = 10 GHz.
3. Operating in a 50 ohm system.
4. Substrate relative permittivity 2.23 and thickness 0.234 mm.

Compare these values with those applying to a simple parallel-coupled microstrip


coupler, and comment on the practical implications.
Signal Transmission, Network Methods and Impedance Matching 163

Input P1 dB below input


(a) Physical layout λgm /4
No
output P2 dB below input
(isolated) λgm /4 λgm /4 λgm /4

Short circuits

Odd mode
(b) Odd and even
mode equivalent
circuits Open circuits

Even mode

Figure 3.35 The orthogonal branch-line directional coupler

A wide and ever-growing variety of further coupler configurations exists. Amongst these
are: branch-line directional couplers (orthogonal and ring types) and the ‘rat-race’ or ring
hybrid. We indicate some examples of these configurations here, but shall not provide design
routines or expressions.
The basic configuration of the (orthogonal) branch-line directional coupler is shown in
Figure 3.35.
These types of couplers have the following properties:

1. Elements such as chokes or filters (particularly in-line band-stop) can be introduced into
joining branches.
2. In many cases relatively high RF/microwave voltages can be handled (there is no risk of
breakdown across a gap).
3. Tight coupling is easy to achieve – but only over limited bandwidths – typically 50%.
4. Impedance transforming is a further use of these designs.

Ring forms of couplers have been known for many years in waveguide, coaxial and strip-
line configurations. The basic design requirements are similar for microstrip realisation and
the 3 dB branch-line and the ‘rat-race’ or ‘hybrid ring’ are shown in Figures 3.36 and 3.37
respectively.

3.8 Network Methods


Network theory is a formidable aid to the electrical engineer. A thorough grasp of this
subject enables a multitude of applications to be easily tackled. Examples of the wide
applicability of network theory are its use in CAE packages to simulate highly complex and
large-scale circuits, to the analysis of single semiconductor devices.
In the following subsections we concentrate on describing a parameter set, the scattering
or ‘s’ parameters, that is of particular relevance to the microwave circuit designer. The
requirement for s-parameters in high frequency applications and the techniques for convert-
ing s-parameter data into other parameter sets (z, y, h, ABCD, etc.) are also described.
164 Microwave Devices, Circuits and Subsystems for Communications Engineering

Z0
λg/4 λg/4
Z0
2
1 Z0 Z0
3
Z0
Z0
2
λg/4 λg/4
Z0
4

Figure 3.36 The 3 dB ring-form branch-line directional coupler

output
(2)
Z0 (3)
Z0 λg/4

λg/4 λg/4
(1)
input Z0 Z0 output
(4)

Z0 2
3λg/4

Figure 3.37 The ‘rat-race’ or ‘hybrid ring’ coupler

3.8.1 Revision of z, y, h and ABCD Matrices


A general two-port network is shown in Figure 3.38.
The most common parameter sets used to describe such a network are:

(i) z-parameters

V1 = z11 I1 + z12 I2 (3.166(a))

V2 = z21 I1 + z22 I2 (3.166(b))

(ii) y-parameters

I1 = y11V1 + y12V2 (3.167(a))

I2 = y21V1 + y22V2 (3.167(b))


Signal Transmission, Network Methods and Impedance Matching 165

I1 I2

Two-Port
V1 Port 1 Port 2 V2
Network

Figure 3.38 Two-port network

(iii) h-parameters

V1 = h11I1 + h12V2 (3.168(a))

I1 = h21I1 + h22V2 (3.168(b))

(iv) ABCD parameters

V1 = AV2 − BI2 (3.169(a))

I1 = CV2 − DI2 (3.169(b))

The selection of a particular set is very much influenced by the application. For example,
manufacturers of bipolar transistors usually provide users with the h-parameters of the
device. This set yields the input and output impedance of the transistor (h11, 1/h22), its current
gain (h21) and its reverse voltage gain (h12), which are key parameters in the selection of
a device. In many circuit simulation packages, however, the admittance (or y-) parameters
are used because of the ease with which large circuits can be described and their matrices
manipulated. The choice of a particular set is also influenced by the ease with which the
parameters can be measured in the particular situation of interest. Very often conversion
between one parameter set and another is required to ease the task in hand. This is readily
performed by simple substitution. For example, to determine z11 from the y-parameters, the
measurement condition for z11 is first determined:

V1
z11 = (3.170)
I1 I2 = 0

Then expressions for V1 and I1 are found from the Y-matrix:

I2 − y22V2
V1 = (3.171(a))
y21
y11( I2 − y22V2 )
I1 = + y12V2 (3.171(b))
y21
Finally, terms containing I2 can be cancelled to give the required parameter transformation:

V1 y22
z11 = = (3.172)
I1 I2 = 0 y11 y22 − y12 y21
166 Microwave Devices, Circuits and Subsystems for Communications Engineering

A similar process can be used to convert between any of the parameter sets. Notice that none
of these parameter sets require knowledge of the source or load impedances connected to the
network. Notice also, however, that these parameter sets do require a voltage or a current
to be set to zero. From a measurement point of view, this requires a short or open circuit
termination. As will be seen later, at microwave frequencies these two conditions are difficult
to satisfy. An alternative parameter set that avoids this requirement, and is therefore particularly
useful for microwave applications, is now described.

3.8.2 Definition of Scattering Parameters


Consider a sinusoidal voltage source of internal impedance ZS applying a voltage to a trans-
mission line of characteristic impedance Z0 terminated in a load impedance ZL, as shown in
Figure 3.39.
The application of such a sinusoidal voltage causes a forward voltage and current wave to
propagate towards the load. These have the complex form:

V(x, t) = A exp ( jω t − ψx) (3.173(a))

A
I(x, t) = exp ( jω t − ψx) (3.173(b))
Z0

If the load ZL is perfectly matched to the line impedance Z0, then all of the power incident on
the load will be absorbed by the load. If ZL is not matched to Z0, however, then a wave will
be reflected from the load. The reflected wave, which travels along the line towards the
source, has the complex form:

V(x, t) = B exp ( jω t + ψx) (3.174(a))

B
I(x, t) = − exp ( jω t + ψx) (3.174(b))
Z0

If impedance ZS is equal to Z0, then the source is matched to the line.

Forward wave
ZS ‘a’

Transmission
Line Z 0 ZL

Reflected wave
‘b’

Figure 3.39 Transmission line as a network


Signal Transmission, Network Methods and Impedance Matching 167

The forward and backward travelling waves can be normalised (such that the amplitude of
each term represents the square root of wave power) and combined, i.e.:

V ( x, t ) A B
= exp ( jω t − ψ x ) + exp ( jω t + ψx ) (3.175(a))
Z0 Z0 Z0

A B
I ( x, t ) Z0 = exp ( jω t − ψ x ) − exp ( jω t + ψx ) (3.175(b))
Z0 Z0

The forward wave ‘a’ (see Figure 3.39) can therefore be defined as:

A
a= exp ( jω t − ψ x ) (3.176(a))
Z0

and the reflected wave ‘b’ can be defined as:

B
b= exp ( jω t + ψx ) (3.176(b))
Z0

hence:

V ( x, t )
=a+b (3.177(a))
Z0

and:

I ( x, t ) Z0 = a − b (3.177(b))

Alternatively (but equivalently), a and b can be defined as:

V ( x, t ) + I ( x, t ) Z0
a= (3.178(a))
2 Z0

and:

V ( x, t ) − I ( x, t ) Z0
b= (3.178(b))
2 Z0

If the ratio s = b/a (referred to as the reflection coefficient) is used as a figure of merit, then:

b V ( x, t ) − I ( x, t ) Z0 Z − Z0
s= = = L (3.179)
a V ( x, t ) + I ( x, t ) Z0 ZL + Z0

When ZL is equal to Z0, then s will be zero and we have a matched line. When ZL is a short
circuit (ZL = 0), then s will be unity with a phase angle of −180°. When ZL is infinity (open
circuit), s will again be unity, but its phase angle is 0°. Between these extreme values of ZL,
the forward and reflected waves interact to form a standing wave. s therefore indicates the
amount of mismatch present in the system. Maximum power is injected into the line when ZS
168 Microwave Devices, Circuits and Subsystems for Communications Engineering

a Z0 I
one-port
Network

Vg V ZL

Figure 3.40 One-port network

and ZL are equal to the impedance Z0. In this fully matched state, the voltage/current ratio
will be constant along the line and equal to Z0. The scattering parameters are based on s, i.e.
the interaction between the forward and reflected waves.

3.8.3 S-Parameters for One- and Two-Port Networks


The one-port problem has been described in the context of a transmission line in the pre-
vious section. Here we relate the solutions to the more general one-port device shown in
Figure 3.40, where Z0 is the output impedance of the signal source (generating voltage Vg)
and ZL is the device (or network) input impedance, and extend the s-parameter description
to two-port networks.
The forward wave (a) and the reflected wave (b) are given by:

V + IZ0
a= (3.180(a))
2 Z0

and:
V − IZ0
b= (3.180(b))
2 Z0

(Note that the explicit argument notation x, t has been dropped but still applies.) The one-
port reflection coefficient or s-parameter is given by Equation (3.179), repeated here:

b ZL − Z0
s= = (3.181)
a ZL + Z0

For a two-port network, the same idea can be applied. For this case, the ‘a’ and ‘b’ variables
are defined for port 1 as:

V1 + Z0 I1
a1 = (3.182(a))
2 Z0

V1 − Z0 I1
b1 = (3.182(b))
2 Z0
Signal Transmission, Network Methods and Impedance Matching 169

and for port 2 as:

V2 + Z0 I2
a2 = (3.183(a))
2 Z0

V2 − Z0 I2
b2 = (3.183(b))
2 Z0

The wave (b1) travelling away from port 1 towards the source is given by the wave incident
on port 1 times the reflection coefficient for port 1 (s11) plus the wave incident on port 2
times the transmission factor from port 2 to port 1 (s12). (The two-port parameter s11 is
therefore the equivalent of s for the one-port network.) Similarly, the wave (b2) travelling
away from port 2 towards the load is given by the wave incident on port 2 times the
reflection coefficient for port 2 (s22) plus the wave incident on port 1 times the transmission
factor from port 1 to port 2 (s21). The full two-port s-parameter equations therefore have
the form:

b1 = s11a1 + s12a2 (3.184(a))

b2 = s21a1 + s22a2 (3.184(b))

The input (port 1) reflection coefficient is therefore given by:

b1
s11 =
a1 a2 = 0
V1
− Z0
(V1 − Z0 I1 )/ 2 Z0 I
= = 1
(V1 + Z0 I1 )/ 2 Z0 V1
+ Z0
I1
Zin − Z0
= (3.185)
Zin − Z0

Note the requirement for a2 = 0 to yield s11 explicitly. This condition is satisfied when port
2 of the network is terminated in Z 0 as shown in Figure 3.41.

a1 Z0 a2
I1 I2

two-port
Vg V1 V2 Z0
b1 Network b2

Figure 3.41 Two-port network terminated at port 2 in Z0


170 Microwave Devices, Circuits and Subsystems for Communications Engineering

The same condition (a2 = 0) is also needed to yield s21 explicitly:

b2
s21 =
a1 a2 = 0
(V2 − Z0 I2 )/ 2 Z0 V − Z0 I2
= = 2
(V1 + Z0 I1 ) 2 Z0
/ V1 + Z0 I1
2V2
= (3.186(a))
Vg
Therefore:

4 | V2 |2
| s21 |2 =
| Vg |2
Power delivered to port 2
= (3.186(b))
Power delivered by Vg

s21 therefore represents the forward transmission coefficient of the network.


s22 and s12 are determined in a similar manner but now the signal generator is applied as
port 2 and port 1 is terminated in Z0 as shown in Figure 3.42.
In this mode a1 is set to zero since port 1 is terminated in Z0. For this case, s22 and s12 are
determined as:

b2
s22 =
a2 a1 = 0
V2 − Z0 I2 Z − Z0
= = out (3.187)
V2 + Z0 I2 Zout + Z0

b1
s12 =
a2 a1 = 0
V1 − Z0 I1 2V
= = 1 (3.188(a))
V2 + Z0 I2 Vg

a1 a2
I1 I2 Z0

two-port
Z0 V1 V2 Vg
b1 Network b2

Figure 3.42 Two-port network terminated at port 1 in Z0


Signal Transmission, Network Methods and Impedance Matching 171

Therefore:

4 | V1 |2 Power delivered to port 1


| s12 |2 = = (3.188(b))
| Vg |2
Power delivered from Vg

s22 therefore represents the reflection coefficient of port 2 and s12 is the reverse transmission
coefficient. s11 and s22 relate to the input and output impedances of the network, and s21 and
s12 relate to the forward and reverse gains of the network.
It is clear that all the s-parameters can be determined by terminating the input or output
ports of the network in Z0. In modern network analysers, the reference impedance (usually
50 ohm) can be specified very accurately over a broad frequency band. As will be seen in
Section 3.10.2, however, the calibration (error correcting techniques) needed to improve the
accuracy of the measured data requires the use of open and short circuit terminations. These
can also be specified quite accurately over a broad frequency band. If this is the case, why
are the s-parameters preferred over other parameter sets for microwave work? The answer is
considered in the following section.

3.8.4 Advantages of S-Parameters


The acquisition of z, y, h and ABCD parameters requires either a short or open circuit
termination. The difficulties of achieving this at microwave frequencies is often quoted
as being their main disadvantage. It should be stressed, however, that this difficulty is in
ensuring that the terminals of the device or circuit are open or short circuit. Broadband open-
and short-circuit reference terminations for calibration purposes are available in a variety
of connector styles and are used for s-parameter calibration purposes (see Section 3.10.2).
The use of such reference terminations would not, however, satisfy the required criteria for
the above parameter sets because of the stray parasitic effects introduced by the terminals
(probably a microstrip arrangement) of the circuit under test.
Even if a true short or open condition could be established, such a termination can in
many cases cause instability. This is true of microwave semiconductor devices having a high
bandwidth. Such devices need to be terminated in some (known) intermediate impedance
(usually 50 ohm) for stability. A further measurement disadvantage of the above parameters
is that the voltages and currents on which they depend vary along a transmission line. It is
also true that the parasitic components of the probe used to measure the voltage and current
can affect the device or network behaviour.
The s-parameter approach avoids all these measurement problems. The word measurement
needs to be stressed here, since all of the disadvantages quoted for z, h, y and ABCD matrices
are measurement-related. It is often the case that the s-parameters data are converted into z,
y, h or ABCD form for analysis purposes. The converted information can be trusted since
the accuracy of the original measured data will be high. The procedure for the conversion of
s-parameter data into z-parameter format is discussed in the next section.

3.8.5 Conversion of S-Parameters into Z-Parameters


The conversion between s-parameters and other parameter sets and vice versa is performed
using the same basic approach as illustrated previously. For this case, however, the load and
source impedance (both assumed to be Z0) must be taken into account.
172 Microwave Devices, Circuits and Subsystems for Communications Engineering

The terminal currents and voltages are:


a1 − b1
I1 = (3.189(a))
Z0
V1 = Z0 (a1 + b1 ) (3.189(b))
where subscripts refer to port number, a refers to incident waves and b refers to reflected waves.
a2 − b2
I2 = (3.190(a))
Z0
V2 = Z0 (a2 + b2 ) (3.190(b))
For the z-parameter set, the two-port equations are:
V1 = z11 I1 + z12 I2 (3.191(a))
V2 = z21 I1 + z22 I2 (3.191(b))
which can be rewritten as:
Z0(a1 + b1) = z11(a1 − b1) + z12(a2 − b2) (3.192(a))
and:
Z0(a2 + b2) = z21(a1 − b1) + z22(a2 − b2) (3.192(b))
Solving for b1 and b2 gives:
z12 (a2 − b2 ) + a1( z11 − Z0 )
b1 = (3.193(a))
Z0 + z11
z21(a1 − b1 ) + a2 ( z22 − Z0 )
b2 =
Z0 + z22
2a1Z0 z21 + a2 [( Z0 + z11 )( z22 + Z0 ) − z21z12 ]
= (3.193(b))
( Z0 + z11 )( z22 + Z0 ) − z21z12
Therefore:
b2
s21 =
a1 a2 = 0
2 Z0 z21
= (3.194(a))
( Z0 + z11 )( Z0 + z22 ) − z21z12
and:
b2
s22 =
a2 a1 = 0
( Z0 + z11 )( z22 + Z0 ) − z21z12
= (3.194(b))
( Z0 + z11 )( Z0 + z22 ) − z21z12
Signal Transmission, Network Methods and Impedance Matching 173

Similarly:

2a2 Z0 z12 + a1[( Z0 + z22 )( z11 − Z0 ) − z21z12 ]


b1 = (3.195)
( Z0 + z11 )( z22 + Z0 ) − z21z12

Therefore:

b1
s12 = (3.196(a))
a2 a1 = 0
2 Z0 z12
= (3.196(a))
( Z0 + z11 )( Z0 + z22 ) − z21z12

and:

b1
s11 =
a1 a2 = 0
( Z0 + z22 )( z11 − Z0 ) − z21z12
= (3.196(b))
( Z0 + z11 )( Z0 + z22 ) − z21z12

Clearly, this procedure is reversed to provide the z-parameters from the s-parameters:

b1 = s11 a1 + s12 a2 (3.197(a))

b2 = s21 a1 + s22 a2 (3.197(b))

Using the definition for a1, a2, b1, and b2, these equations can be rewritten as:

V1 − Z0 I1 = s11(V1 + Z0 I1) + s12(V2 + Z0 I2) (3.198(a))

and:

V2 − Z0 I2 = s21(V1 + Z0 I1) + s22(V2 + Z0 I2) (3.198(b))

Solving for V2:

Z0{2s21 I1 + I2 [(1 − s11 )(1 + s22 ) + s12 s21 ]}


V2 = (3.199)
(1 − s22 )(1 − s11 ) − s12 s21

Therefore:

V2
z22 =
I2 I1 = 0
Z0 [(1 − s11 )(1 + s22 ) + s12 s21 ]
= (3.200(a))
(1 − s22 )(1 − s11 ) − s12 s21
174 Microwave Devices, Circuits and Subsystems for Communications Engineering

and:

V2
z21 =
I1 I2 = 0
2 Z0 s21 (3.200(b))
=
(1 − s22 )(1 − s11 ) − s12 s21
Similarly,

Z0{I1[(1 − s22 )(1 + s11 ) + s12 s21 ] + 2s12 I2}


V1 = (3.201)
(1 − s22 )(1 − s11 ) − s12 s21

Therefore:

V1
z11 =
I1 I2 = 0
Z0 [(1 − s22 )(1 + s11 ) + s12 s21 ]
= (3.202(a))
(1 − s22 )(1 − s11 ) − s12 s21

and:

V1
z12 =
I2 I1 = 0
2 Z0 s12
= (3.202(b))
(1 − s22 )(1 − s11 ) − s12 s21

The same procedure can be adopted to convert the s-parameters into y, h or ABCD parameters
and vice versa.

3.8.6 Non-Equal Complex Source and Load Impedances


In the above treatment of scattering parameters, it has been assumed that the impedance at
ports 1 and 2 are equal (Z0). This is because network analysers usually have both ports
terminated in the same impedance. However, in many circuit applications, this condition
may not apply with different impedances at port 1 (Z01) and at port 2 (Z02). It is also true, in
general, that the terminating impedances will be complex. Although the s-parameter approach
is still valid under these conditions, the expressions defining the forward and reflected waves
need to be modified. For port 1 they become:

V1 + Z01 I1
a1 = (3.203(a))
[2( Z01 + Z01
* )]1/2

V1 − Z *01 I1
b1 = (3.203(b))
[2( Z01 + Z01
* )]1/2
Signal Transmission, Network Methods and Impedance Matching 175

and for port 2 they become:

V2 + Z02 I2
a2 = (3.204(a))
[2( Z02 + Z02
* )]1/2
V2 − Z02
* I2
b2 = (3.204(b))
[2( Z02 + Z02
* )]1/2
where * denotes the complex conjugate. The conversion between s- and z- (or y, h or ABCD)
parameters clearly needs to consider these modified expressions. If the previously outlined
approach is followed, then the following equations result for the z-parameters in terms of the
s-parameters:

* + s11Z01 )(1 − s22 ) + s12 s21Z01


( Z 01
z11 = (3.205(a))
(1 − s22 )(1 − s11 ) − s12 s21

2s12 ( R01 R02 )1/2


z12 = (3.205(b))
(1 − s22 )(1 − s11 ) − s12 s21

2s21( R01 R02 )1/2


z21 = (3.205(c))
(1 − s22 )(1 − s11 ) − s12 s21

* + s22 Z02 )(1 − s11 ) + s12 s21Z02


( Z 02
z22 = (3.205(d))
(1 − s22 )(1 − s11 ) − s12 s21

In these expressions, R01 and R02 represent the real component of Z01 and Z02 respectively.
The expressions defining the s-parameters in terms of the z-parameters are:

( z11 − Z01
* )( z22 − Z02 ) + z12 z21
s11 = (3.206(a))
( z11 + Z01 )( z22 + Z02 ) − z12 z21

2 z12 ( R01 R02 )1/2


s12 = (3.206(b))
( z11 + Z01 )( z22 + Z02 ) − z12 z21

2 z21( R01 R02 )1/2


s21 = (3.206(c))
( z11 + Z01 )( z22 + Z02 ) − z12 z21

and:

( z11 − Z01 )( z22 − Z *02 ) − z12 z21


s22 = (3.206(d))
( z11 + Z01 )( z22 + Z02 ) − z12 z21

Self-assessment Problem
3.38 Verify these results by following the procedure described previously. Repeat for
the other parameter sets.
176 Microwave Devices, Circuits and Subsystems for Communications Engineering

3.9 Impedance Matching


An important task for the RF engineer is impedance compensation or, as is more commonly
known, matching. This refers to consideration, in the design process, of the impedance prop-
erties of the particular subsystem in order to ensure it is compatible with its environment.
Impedance matching arises as a consequence of the power transfer theorem, which points
out that a badly mismatched design will result in large power losses.
The need for close control of impedance is more general, however, than simply considera-
tions of power transfer. For several applications, rather than perfect matching (for maximum
power transfer), the requirement is for controlled mismatch. When an amplifier is designed,
for example, the selection of the input and output matching networks not only provides for
good power transfer, but also determines other characteristics of the design such as gain and
noise figure.
Impedance matching may be relatively straightforward for narrow-band designs but the
complexity increases with increasing bandwidth. The difficulty arises from variation with
frequency in the network properties. Complex circuits are therefore required when broadband
designs are implemented.

3.9.1 The Smith Chart


Tedious mathematical calculations are required if the basic transmission line equations are
used to provide solutions to even simple problems. In the early days of RF engineering
replacement of these calculations with a faster, graphical, method in order to both accelerate
and simplify the design process was therefore very desirable. With the advent of CAD
software the requirement for graphical methods has become less essential from a practical
point of view. The insight that a geometrical representation of impedance problems (and their
solution) yields, however, has meant that the most popular of these methods, a chart devised
by Philip H. Smith, has become part of the standard language of the RF designer.
The Smith chart provides a graphical tool that allows quick and accurate manipulation of
impedance data. One of its most useful applications is the design of matching networks although
it can also be applied to solve many other problems associated with transmission lines.
The chart is developed via a transformation of a Cartesian coordinate representation
of normalised impedance to a polar coordinate representation. Lines of constant normalised
resistance, r, and normalised reactance, x, on the Cartesian plane are transformed to circles
on the polar plane.
To gain an insight into the development of the Smith chart, consider the equation govern-
ing the reflection coefficient at the load:
ZJ − Z0
ΓJ = = | ΓJ | e jϑ = Γx + j Γy
J
(3.207)
ZJ + Z0
where Zj is the load impedance. Assuming that the load reflection coefficient is | Γj | ≤ 1,
indicative of a passive load, then Γj must lie within or on the unit radius circle. Furthermore,
the reflection coefficient at a distance d from the end of the line (having a characteristic
impedance of Z0) will also lie within the unity circle since:
Γd = | ΓJ | e −2α de j (ϑ J −2 β d ) = | Γd | e j (ϑ J −2 β d ) (3.208)
where α and β are attenuation and phase constants respectively.
Signal Transmission, Network Methods and Impedance Matching 177

Γy

Γ =1

Γj

ϑj ϑj − 2βd
Γx

Γd

Figure 3.43 Circles of constant reflection coefficient, | Γ |

In simple terms Equation (3.208) means that for a length d the (complex) reflection
coefficient will be rotated by an angle of −2βd radians (Figure 3.43).
From Equation (3.207) we can derive the normalised impedance in terms of the reflection
coefficient:
Z 1 + ΓJ e −2γ d
z= = (3.209)
Z0 1 − ΓJ e −2γ d
where γ = α + jβ is the propagation constant. For the purpose of derivation we can assume
that the impedance is given at the load, i.e. d = 0. This does not reduce the generality of the
argument, however, since any given point of the line can be simulated by a different phy-
sical load at the same position:
1 + ΓJ R + jXJ
z= = J = r + jx (3.210(a))
1 − ΓJ Z0
also:
z −1
ΓJ = = Γx + j Γy (3.210(b))
z +1
178 Microwave Devices, Circuits and Subsystems for Communications Engineering

Γy

Γ =1

r=0
r=0.4
r=0.5
r=1
r=2
Γx

Figure 3.44 Circles of constant normalised resistance, r

Γy

Γ =1

+jx x=1
x=0.5 x=2

Γx
x=0

-jx

Figure 3.45 Circles of constant normalised reactance, x


Signal Transmission, Network Methods and Impedance Matching 179

From manipulation of these two equations we have:

1 − Γ x2 − Γ y2
r= (3.211(a))
(1 − Γx )2 + Γ y2

and:

2Γy
x= (3.211(b))
(1 − Γx )2 + Γ y2

Rearranging the previous pair of equations we arrive at the following:


2 2
 r   1 
Γx − + Γ y2 = (3.212(a))
 1 + r 1 + r
2 2
 1  1
(Γx − 1) 2 + Γy − = (3.212(b))
 x  x

The curves described by Equations (3.212(a)) and (3.212(b)) on the Γ-plane represent a
family of circles. Equation (3.212(a)) describes the locus of points having constant r whereas
Equation (3.212(b)) describes the locus of points having a constant x. These two sets of
circles are shown in Figures 3.44 and 3.45.
The geometrical parameters of the curves are given explicitly in Table 3.2:

Table 3.2 Centre coordinates and radius of circles of constant r and x

Centre Radius

 r  1
Circles of constant r  , 0
1 + r  1+r
 1 1
Circles of constant x 1, 
 x x

The superposition of these two families of curves results in the Smith chart shown in
Figure 3.46.
Although the constant Γ circles (Figure 3.43) are not usually explicitly shown on the
Smith chart, they are implied, i.e. the reflection coefficient plane in radial co-ordinates is
implicitly superimposed on the normalised impedance plane. Thus any point on the chart
can be read as an impedance or as a reflection coefficient.
The important properties of the Smith chart (see Figure 3.47) can be summarised as follows:

1. It represents all passive impedances on a grid of constant r and x circles.


2. It contains the corresponding reflection coefficients in polar co-ordinates; the angle being
read on the peripheral scale and the magnitude being calculated using:
180 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 3.46 Smith chart

radius of circle on which point is located


|Γ| =
radius of the chart

3. The upper half of the diagram represents positive reactance values (inductive elements).
4. The lower half of the diagram represents negative reactance values (capacitive elements).
Signal Transmission, Network Methods and Impedance Matching 181

0.5 Inductive region 2


Vmax, Imin, SWR

SWR circle
Γ ∠Γ
gen

0 1 ∞
load

Vmin, Imax
0.5 Capacitive region 2

Figure 3.47 Properties of the Smith chart

5. The arc around the Smith chart corresponds to a movement of half a wavelength (λ /2);
any other movement across an arc between any two points can be read on the chart peri-
phery in terms of wavelengths.
6. The horizontal radius to the left of the centre corresponds to voltage minimum and
current maximum (Vmin, Imax).
7. The horizontal radius to the right of the centre corresponds to the standing wave ratio
(SWR), the voltage maximum and the current minimum (Vmax, Imin).
8. By rotating the reflection coefficient plane by 180° the corresponding quantities are read
as admittances. The transformation is Γ → −Γ, in which case:

1
−1
 z − 1 1 − z y −1
−Γ = − = = z = (3.213)
 z + 1 1 + z 1 y +1
+1
z

i.e., when Γ → −Γ then z → y. An admittance chart can therefore be created by simply


rotating the impedance chart through 180°. This transformation is shown in Figure 3.48.

A variation on the Smith chart called the immittance chart (impedance admittance) is
generated if we superimpose the impedance and the admittance charts (Figure 3.49). The
advantage is that we can now read each point as admittance or impedance simply by reading
values from the different grids, thus making the impedance to admittance transformation
easier.
182 Microwave Devices, Circuits and Subsystems for Communications Engineering

0.5 2

gen
0 ∞
load

0.5 2

Figure 3.48 Impedance – admittance transformation

Finally, we can generalise the Smith chart to include active elements, by permitting
values for the reflection coefficient that are larger than unity. The result, called the com-
pressed Smith chart, is shown, schematically, in Figure 3.50. This form is especially useful
for the design of oscillators, where negative resistances are common. In the usual format it
allows gain of 10 dB, which corresponds to reflection coefficient magnitude less than 3.16.
All the properties of the normal Smith chart hold for the compressed Smith chart.

3.9.2 Matching Using the Smith Chart


The single most common use of the Smith chart is matching a load-impedance to a source-
impedance.

3.9.2.1 Lumped element matching


There are several networks that provide impedance transformation but the simplest approach
is to connect a series or shunt lumped component to the load thereby changing the input
impedance. To evaluate the effect of a lumped element connected to the load we can
examine the following cases.

1. Series connection of ideal inductor or capacitor

The connection of a series element, Zel, to an arbitrary load, ZL, (Figure 3.51) will alter the
overall input impedance, Zin, of the network.
Since the component is introduced in series the input impedance is given by:

Zin = Zel + ZL (3.214)


Signal Transmission, Network Methods and Impedance Matching 183

Figure 3.49 Immitance chart

If the component introduced is purely imaginary, i.e. Zel = jXel, then the Smith chart descrip-
tion of the effect is movement along a circle of constant r. The sense of the movement is
determined by the sign of the reactance introduced. An inductor (positive reactance) changes
the overall impedance towards the upper half of the chart, whereas a capacitor (negative
reactance) changes it towards the lower half of the chart, Figure 3.52.
184 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 3.50 Compressed Smith chart

Zel

ZL
Zin

Figure 3.51 Series connection of a lumped element

2. Shunt connection of ideal inductor or capacitor

It is easier to follow the effect of a shunt element, Figure 3.53, using the admittance chart.
The input admittance for the parallel network is given by:

Yin = Yel + YL (3.215)

If the component introduced is purely imaginary, i.e. Yel = jBel, then the Smith chart descrip-
tion is movement along a circle of constant (normalised conductance) g. The sense of the
movement is determined by the sign of the susceptance introduced. A capacitor (positive
susceptance) changes the overall admittance towards the lower half of the (admittance)
chart, whereas an inductor (negative suseptance) changes it towards the upper half of the
chart, Figure 3.52.
A combination of series and shunt elements can be used to match an arbitrary load to the
source (at a single frequency). The most convenient tool for multi-element matching is the
Signal Transmission, Network Methods and Impedance Matching 185

0.5 2

Series Shunt
Inductor Inductor

0 ∞

Series Shunt
Capacitor Capacitor

0.5 2

Figure 3.52 Effect of series element on impedance and shunt elements on admittance

Zel

Zin ZL

Figure 3.53 Shunt connection of lumped element

immittance chart, since it allows rapid transformation between the admittance and impedance.
The general technique (illustrated in Figure 3.53) is summarised in the following steps:

1. Normalise the load impedance, ZL, to zL = ZL /Z0 and plot this on the immittance chart
2. Determine the admittance of the matching element using (a) or (b) below:

(a) If the last element of the matching network is a parallel component, then the admittance
chart is used, in which case movement from zL is along a circle of constant g (adding
susceptance) until the unit normalised resistance (r = 1) circle is intersected. This point
gives the required normalised input admittance yin. The normalised admittance, yel, of
the required element is then determined from the normalised version of Equation
(3.215).
186 Microwave Devices, Circuits and Subsystems for Communications Engineering

(b) If the last element of the matching network is a series component, then the impedance
chart is used, in which case movement is along a circle of constant r (adding reactance)
until the unit normalised conductance (g = 1) circle is intersected. This point gives
the required normalised input impedance, zin. The normalised impedance, zel, of the
required series element is then determined from the normalised version of Equation
(3.214).

3. Determine the admittance of the second element from the new impedance value (on either
of the g = 1 or the r = 1 circle) by following the previous procedure.
4. Determine the inductance and capacitance of the components from their impedance and
admittance values. The de-normalised values are, for a series capacitor, series inductor,
parallel capacitor and parallel inductor, given respectively by:

1
C= (3.216(a))
ω xZ0

xZ 0
L= (3.216(b))
ω

b
C= (3.216(c))
ω Z0

Z0
L= (3.216(d))
ωb
where Z0 is the impedance used to normalise the Smith chart (the value corresponding to
the centre of the chart), x is the normalised reactance and b is the normalised susceptance.

0.5 2

0 1

ZL

0.5 2

Figure 3.54 Matching using lumped elements


Signal Transmission, Network Methods and Impedance Matching 187

The value selected for Z0 can be different from the source impedance. When lumped elements
are used to provide matching we normally choose a convenient value that brings all the
impedances used towards the centre of the chart. (Theoretically this value can be selected to
be complex but the difficulty involved in manipulating complex numbers usually means that
a real value is chosen.)

3.9.2.2 Distributed element matching


As the operational frequency increases, it becomes correspondingly difficult to employ
lumped components. The alternative is to incorporate distributed elements in the form
of transmission line segments. A transmission line can exhibit inductive or capacitive
behaviour depending on its length, characteristic impedance and terminal load. This property
can be exploited in order to design matching networks. There are several techniques for
the design of such distributed matching networks. The most common ones are presented
next.

3.9.2.3 Single stub matching


The distributed element equivalent of the lumped element L-network consists of a length
of transmission line connected directly to the load (corresponding to the series component)
and a transmission line stub connected in parallel to the line (Figure 3.55). The principle
behind this configuration is that any length of line moves the admittance of the load along
an arc of a circle of constant reflection coefficient. After a certain rotation (corresponding
to the required transmission line length), the real part of the admittance will be equal to
that of the source. At that point the introduction of a susceptive load will provide the
required matching. This can be provided by a lumped inductor or capacitor but it is more
common in practice to incorporate the susceptive properties of a short circuited or open
circuited line.

1
jd

ynl yj
ys

Zo ZL
jI

Figure 3.55 Single stub matching network


188 Microwave Devices, Circuits and Subsystems for Communications Engineering

jd
1

0.5 2
B
gen C

SC
0 ∞
OC 1
load

jI
A
0.5 2
D

Figure 3.56 Single stub matching

The design steps, illustrated in Figure 3.56, are summarised as follows.

1. Normalise the impedance of the load to the characteristic impedance of the line and plot
this on the Smith chart (point A).
2. Translate the impedance to admittance (point B), after which the Smith chart should be
read as an admittance chart.
3. Move the reference plane towards the generator along a constant reflection coefficient
circle until the unit conductance circle (g = 1) is intercepted (point C). The arc between
B and C determines the length (in wavelengths) of the series line.
4. The susceptance of the stub introduced at point C should cancel the existing susceptance
at that position. At the connection point therefore:
yn1 = yj + yS (3.217(a))
i.e.:
yS = yn1 − yj (3.217(b))
where yS is the normalised admittance of the stub, yn1 is the normalised admittance after
the stub has been added (usually equal to the normalised source admittance), and yj is
the admittance looking towards the load at the point of the introduction of the series line,
see Figure 3.55.
5. Determine the admittance yS (which is purely imaginary) on the perimeter of the Smith
chart (point D). The required length of the stub (jI) is provided by the arc between
point D and the short or open circuit termination (SC or OC on the Smith chart) in the
direction of the load.
Signal Transmission, Network Methods and Impedance Matching 189

Admittance has been employed since it is easier to use for parallel elements, distributed or
lumped. A series stub can, in principal, provide the same effect although this solution is
rarely used in practice. Nevertheless the design process remains the same apart from the fact
that the impedance chart is used instead of the admittance chart.
The selection of an open-circuit or short-circuit stub depends on the kind of transmission
line. When coaxial lines are employed, for example, it is easier to implement a short circuit
whereas in microstrip it is easier to implement an open circuit.
The source impedance here, as in most practical cases, is assumed to be equal to the
characteristic impedance of the line. In the case where the source and the characteristic
impedance are different, the normalising impedance is still that of the line (Z0). The design
process varies, but only slightly, since the source impedance does not coincide with the
centre of the chart. The basic steps, however, are the same.

3.9.2.4 Double stub matching


Although any impedance can be matched with a single stub, it is necessary to vary (i.e.
select freely) both the stub’s position and length. This is not always easy to achieve and in
some applications the position of the stub is fixed. In this case two or three stubs in fixed
positions from the load can be used to provide the matching.
The double stub configuration is shown in Figure 3.57. It consists of two short-circuit stubs
connected in parallel at fixed positions on the line. (The stubs may also be open circuit or a
combination of open and short circuit.) The distance between the stubs (jd2) is usually selected
to be 1/8, 3/8, 5/8 of a wavelength, whereas the position of the first stub from the load (jd1)
depends on the region (of the Smith chart) in which the load is most likely to be found.
The design process, illustrated in Figure 3.58, proceeds as follows:

1. Normalise the load impedance to the characteristic impedance of the line (ZO) and plot
it on the Smith chart.
2. Transfer the normalised impedance to the admittance chart (because the stubs are con-
nected in parallel and it is, therefore, easier to manipulate admittances).

2 1
jd2 jd1

yn2 yd2 yn1 yd1


y stI

y stI
I

St St
ub ub
II I
Zo jI ZL
jII

Figure 3.57 Double stub matching network


190 Microwave Devices, Circuits and Subsystems for Communications Engineering

jd2 1

0.5 2

gen D′

yL
0 1 ∞
zL I
C
load

B
jd1
0.5 C′ 2

Figure 3.58 Double stub matching

3. Rotate the unity conductance circle (g = 1) towards the load by an amount corresponding
to the length jd2, thus determining the spacing circle (circle I).
4. Move the normalised load admittance (yL) along a constant reflection coefficient circle
towards the generator by an amount corresponding to jd1 and read the admittance yd1 at
point B.
5. Determine the intersections of the constant conductance circle that contains point B and
the spacing circle (points C and C′), and read the admittances yn1 and y′n1.
6. Calculate the susceptance to be introduced by the stub I from:

ystI = yn1 − yd1 (3.218(a))

y′stI = y′n1 − yd1 (3.218(b))

7. Determine the lengths jI, j′I from the chart’s circumferential arc in the direction of the
load between the admittance ystI and y′stI and the short circuit (or open circuit if appropriate)
termination of the stub.
8. Rotate points C and C′ towards the generator, along a constant reflection coefficient
circle, determining points D and D′ on the g = 1 circle. These points correspond to the
admittances yd2 and y′d2 just before stub II.
9. Calculate the susceptance to be introduced by stub II from:

ystII = yn2 − yd2 (3.219(a))

y′stII = yn2 − y′d2 (3.219(b))


Signal Transmission, Network Methods and Impedance Matching 191

jd2 1
jd1
0.5 2
III

II
gen

0 1 ∞
I
load

0.5 2

Figure 3.59 Load admittancies that cannot be matched using double stub

10. Finally determine lengths jII, j′II from the chart’s circumferential arc in the direction
of the load between the admittance ystII and y′stII and the short circuit (or open circuit if
appropriate) termination of the stub.

In general, there are two solutions for a given load. Admittances exist, however, that cannot
be matched with a given stub position. No yn1 admittances falling within circle II (tangent
to the spacing circle), see Figure 3.59, can be matched. For those admittances that lie on
the circumference of circle II, only one solution exists. The loads for which matching cannot
be achieved are those lying within circle III, determined by moving circle II by jd2 towards
the load.
To solve the ‘no solution’ problem an additional stub can be employed at the load
(Figure 3.60) to move the (composite) load admittance outside the critical area.

3.9.3 Introduction to Broadband Matching


The term broadband matching loosely implies that impedance compensation should be
achieved over frequency ranges larger than 50%. The design difficulty in this case arises
from the large variations of the load impedance across the matching frequency band. When
such frequency-dependent loads (e.g. transistors or broadband antennas) require matching,
the design aims usually shift from achieving perfect matching to improving the broadband
performance. This is usually possible to achieve when some maximum permissible SWR
is acceptable, thus enabling acceptable reflection coefficient over the whole frequency range
while allowing some frequency ripple. In general, the larger the required bandwidth, the
192 Microwave Devices, Circuits and Subsystems for Communications Engineering

3 2 1
jd2 jd1

St St St
ub ub ub
III II I
ZL jI
Zo jIII jII

Figure 3.60 Triple stub matching network

Z1 Z2

Figure 3.61 Layout of a line transformer

greater must be the allowed matching ripple and the more modest the guaranteed matching
performance at a particular frequency.
Distributed elements usually have poor broadband performance due to their fixed
geometric characteristics although some distributed networks exhibit better broadband per-
formance than others. A line transformer (Figure 3.61), for example, allows better broadband
matching than a single stub, which in turn is more broadband than a double stub. Generally
speaking, the larger and the more often, abrupt changes in characteristic impedance are
encountered, the smaller the operational bandwidth of the design.
In practice when broadband performance is needed, it is usual to use multi- (more than
three) element designs. The advantage of these designs is that they can determine the overall
quality factor (Q) of the network, the lower the Q, the more broadband the design. The
Smith chart is a valuable tool in this case as well. The Q of a series impedance circuit is the
ratio of the reactance to the series resistance. Each point on the Smith chart therefore has a
Q associated with it. The locus of impedances on the chart with equal Q is a circular arc that
passes through the open circuit (O/C) and short circuit (S/C) load points. Several Q circles
are shown in Figure 3.62.
Constant Q circles can be used to provide limits within which the matching network
should remain in order to provide a required operational bandwidth. The design process,
illustrated in Figure 3.63, is as follows.

1. Normalise the source and load impedances with a convenient value (Z0) and plot them on
the Smith chart.
2. Plot the Q circle that corresponds to the overall quality factor required (a low value,
usually less than 5 is normally selected).
Signal Transmission, Network Methods and Impedance Matching 193

0.5 2
Q circles

0 1

0.5 2

Figure 3.62 Constant Q circles

0.5 2

Q=1

0 ZL 1

0.5 2

Figure 3.63 Broadband matching using lumped components

3. If the matching network is a π-network (e.g. the last element is a parallel component)
move from the load along a constant g-circle on the immittance chart until the Q-circle is
reached. If the matching network is a T-network (e.g. the last element is a series component)
move from the load along a constant r-circle until the Q-circle is reached.
4. Introduce an element of the opposite sign from the first one until the real axis is reached.
5. Repeat the previous process introducing as many elements as required until the source is
reached.
194 Microwave Devices, Circuits and Subsystems for Communications Engineering

When a, lossless, two-element L-section is used for matching, the Q is automatically limited
by the source and load impedance:

RP
QS = QP = −1
RS (3.220)

where RP is the parallel resistance and RS is the series resistance of the L-section. A section
of at least three elements should, therefore, be used when broadband matching is required.

3.9.4 Matching Using the Quarter Wavelength Line Transformer


The input impedance, Zg, of a λ /4 line with characteristic impedance ZT terminated in load
ZL is given by:

Z T2
Zg =
ZL

i.e.:

ZT = Zg ⋅ Z L (3.221)

When both the input and load impedance of the transformer are purely resistive (real), then
a quarter wavelength line having characteristic impedance ZT can be used to match ZL to Zg.
When one or both impedances are reactive, however, then a λ /4 line is not sufficient. In such
cases a single section transformer can be used as described below.

3.9.5 Matching Using the Single Section Transformer


A technique similar to the use of a λ /4 transformer can be employed to match any impedance
point that lies within the unit resistance or the unit conductance circle. A single section line
is incorporated between the load and source impedances (Figure 3.64).
The following design process (illustrated in Figure 3.65) can be used to determine the
required length and characteristic impedance of the line:

1. Normalise ZL by Z0.
2. Connect point A corresponding to zL with the centre of the Smith chart.
3. Find the perpendicular bisector and determine Point C at the intercept with the real axis.

Z0 ZT ZL

Figure 3.64 Single section transformer


Signal Transmission, Network Methods and Impedance Matching 195

0.5 D 2
jT

A
gen

0 SC
OC B C 1 ∞
E
load

jI

0.5 2

Figure 3.65 Matching using a single section line

4. Draw a circle around point C with radius equal to the distance from the centre thus
arriving at point B, corresponding to a de-normalised impedance ZB.
5. Calculate the characteristic impedance of the single section line (ZT) using:

ZT = ZB ⋅ Z0 (3.222)

6. Re-normalise ZL to the calculated value of ZT determining point D.


7. Move towards the generator along a constant reflection coefficient circle until the real
axis is intersected at point E. The length of the line in wavelengths is determined from
the arc lT along the chart circumference.

A quick investigation of the design process identifies the locus of the points that can be
matched using a single section. Point B will lie outside the Smith chart for any load impedance
outside the unit resistance and conductance circles. To extend the method to cover these
impedance values a second section of line is needed to move the load, when required, into
the permitted region.

3.10 Network Analysers


The development of network analysers able to measure accurately and quickly the s-
parameters of a device or circuit over a broad frequency range has made a significant
contribution to the advancement of microwave engineering. Measurements that in the
196 Microwave Devices, Circuits and Subsystems for Communications Engineering

past took days to perform can now be completed in a fraction of the time. In addition,
measurements which in the past would not have been feasible are now common. Examples
of this are measurements on production lines to monitor the performance of circuits or
devices. Computer-controlled instruments enable the information gathered to be stored for
statistical purposes. Test programs can also be generated during the product design cycle,
thereby integrating the design and production stages. In the following subsections we
describe the principles of operation of a typical network analyser, the models and calibration
procedures employed to reduce measurement errors, and the test jigs and probes required for
the characterisation of semiconductor devices.

3.10.1 Principle of Operation


A simplified diagram of a typical vector network analyser is shown in Figure 3.66. The
synthesised signal source generates either a continuous wave or a swept RF signal. The
frequency range of the source depends on the user’s needs. For example, in the characterisa-
tion of gallium-arsnide devices, a 40 GHz source would be required because of the high
frequency ( ft > 25 GHz) capability of these devices. The characterisation of circuits for
mobile communications applications operating at 1.8 GHz, however, would not require such
a high frequency source. The power delivered by the source to the test set is usually levelled
using automatic control. Phase locking is achieved by routing a portion of the RF signal
through the test set to the R input of the receiver. Here, the signal is sampled by a phase
detection circuit and fed back to the source.
The RF signal from the source is applied to the circuit/device under test through the test
set. The signal transmitted through the device or reflected from its input is fed through the
A and B inputs to the receiver. Here the transmitted and reflected signals are compared with
the incident or reference signal at input R.
In the receiver the R, A and B RF inputs are converted or sampled to form a low fre-
quency intermediate frequency (IF). The sampling process retains the magnitude and phase
information of the RF signal. The IF data is usually converted into digital signals using an
analog-to-digital converter (ADC) for further processing.

Phase Locking Loop

R
Synthesised Test A
Receiver
Source Set B

To Device or Output Data


Circuit Under
Test

Figure 3.66 Main circuit blocks of a vector network analyser


Signal Transmission, Network Methods and Impedance Matching 197

RVCO VCO SRD


100 kHz 30–60 MHz

Crystal RF
Osc. Phase RF Source Output
1 MHz Sampler
Comparator (YIG or Cavity)

1st IF

1st IF

Figure 3.67 Simplified schematic of a signal source

In the following subsections each of the network analyser components is described in


more detail.

3.10.1.1 The signal source


A simplified schematic diagram of a possible signal source for a 3 GHz network analyser
is shown in Figure 3.67. The reference voltage controlled oscillator (RVCO) generates a
100 kHz reference signal. This is fed to the VCO which generates a signal in the 30 to
60 MHz frequency range. This signal can be a continuous wave (CW) or swept, and is
synthesised and phase locked to the 100 kHz reference signal.
The signal from the VCO is fed into the step recovery diode (SRD), which multiplies the
incoming fundamental signal into a comb of harmonic frequencies. The harmonics are used
as the first local oscillator signal to the sampler.
The crystal oscillator generates a 1 MHz reference signal, which is fed to the phase com-
paritor. This in turn sets the RF source to a first approximation of the RF signal frequency.
The signal is fed to the sampler and combined with the SRD signal to generate a difference
frequency. This difference frequency is filtered and fed back to the phase comparitor. The
difference between the filtered frequency difference and the 1 MHz reference frequency
generates a voltage that adjusts the RF source frequency closer to the required frequency.
This iterative process continues until the difference frequency is equal to the 1 MHz signal.

3.10.1.2 The two-port test set


Figure 3.68 shows a simplified diagram of a test set for two-port circuit or device testing. The
power splitter diverts a portion of the incoming RF signal to the R input of the receiver and
this acts as the reference signal. The remaining RF input signal is routed through a switch
to one of two bi-directional bridges. The switch enables either forward (S11, S21) or reverse
(S22, S12) measurements to be performed on a two-port network. The position of the switch is
controlled by the network analyser and depends on the measurement specified by the user.
A bias tee is normally included for each port to enable the device under test to be biased.
198 Microwave Devices, Circuits and Subsystems for Communications Engineering

RF input
from source Power R input
Splitter of Receiver

B input
of Receiver

Bidirectional
Port 2
bridge

Bidirectional
Port 1
bridge

A input
of Reveiver

Figure 3.68 Simplified schematic of a two-port test set

3.10.1.3 The receiver


A simplified schematic of a receiver is shown in Figure 3.69. The three sampler and mixed
circuits are the same. The sampler and mixer essentially down-convert the incoming RF
signal to a fixed low (4 kHz) 2nd IF. The amplitude and phase of this IF signal correspond
to those of the incoming RF signal. The 2nd IF produced by the mixer is essentially the
difference between the 1st IF (see source description) and the 2nd reference frequency.
The 2nd IF signal from each sampler/mixer circuit is fed to the multiplexer which directs
each of the signals to the analog-to-digital converter. The resulting digital information is
then processed.
A typical automated or computer-controlled measurement system is shown in Figure 3.70.
The PC controls the DC bias supply (PSU1 and PSU2), the network analyser and the printer/
plotter through the IEEE-488 or GPIB interface bus. The PC enables rapid and complex
measurements to be made on circuits or devices. Since the power supplies are also under
computer control, the s-parameter measurements can be performed as a function of bias. This
would be important in device characterisation, where the non-linearity of the component is
an aspect of interest to the circuit designer.

3.10.2 Calibration Kits and Principles of Error Correction


Prior to any device or circuit measurement, a calibration procedure should be followed to
minimise the effects of the sources of errors inherent in the system. Here the system includes
the network analyser, connecting cables, connectors, test jigs used to accommodate the
device, etc. Although the calibration procedures are built into the network analyser itself, it
is important to understand how they operate so that their limitations are understood.
Signal Transmission, Network Methods and Impedance Matching 199

2nd IF
R input Multiplexer
Sampler/Mixer and
ADC

A input
Sampler/Mixer 2nd IF

B input
Sampler/Mixer 2nd IF

2nd Reference
frequency

Figure 3.69 Receiver configuration

IEEE-488 bus
PC

Plotter
Printer
PSU 1
Network
Analyser
PSU 2

Device or
Circuit under
test

Figure 3.70 Typical equipment configuration for s-parameter measurements

The errors present in the system can be divided into two groups: systematic and random.
Random errors cannot be accounted for by the calibration procedures but systematic errors
are predictable and can therefore be compensated for. The principal systematic errors are:

1. Source mismatch
This is the error caused by the impedance mismatch between port 1 of the network
analyser (the source) and the input of the device under test. Some of the incident signal
will be reflected back and forth between the source and the device under test, a fraction
200 Microwave Devices, Circuits and Subsystems for Communications Engineering

of the ‘error’ signal being transmitted through the device under test to port 2 of the
network analyser. Such a mismatch therefore affects both reflection and transmission
measurements.
2. Load mismatch
This is the error caused by the mismatch between the output impedance of the device
under test and the impedance of port 2 of the network analyser. Some of the incident
signal will be reflected back and forth between port 2 and the device under test, a fraction
of the reflected signal from port 2 being transmitted through the device to port 1. If the
device has a low insertion loss, then the signal reflected back and forth between port 1
and port 2 can cause significant errors.
3. Cross-talk
The coupling of energy between the signal paths of the network analyser affects the quality
of transmission measurements. In modern network analysers, however, the isolation is good
and the largest uncertainty arises from the device under test. The calibration procedures
described later eliminate much of this error.
4. Directivity
Ideally, the directional bridge or coupler in the test set should isolate the forward and
reverse travelling waves completely. In reality, a small amount of incident signal appears
at the coupled output due to leakage. Directivity specifies how well the coupler can separate
the two signals. This error is usually small in modern network analysers, the largest
uncertainty arising from the device under test.
5. Tracking
This is the vector sum of all the previously described sources of error as a function of
frequency, the magnitude and phase of the errors being taken into account. It is usual for
the tracking error to be split into two components: the transmission and reflection tracking
error.

As can be seen, there are a large number of errors that need to be accounted for, the majority
of which are addressed by the error models employed by the network analyser. These are
discussed next, starting with the simplest, one-port, case.
For one-port devices, the measured reflection coefficient (s11M) differs from the actual
(i.e. true) reflection coefficient (s11A) of the device under test because of the errors discussed
above. The source mismatch error (ES), directivity error (ED) and tracking error (ET) are
most important. Knowledge of these errors allows the actual reflection coefficient to be
determined using:

s11A ET
s11M = ED − (3.223)
1 − ES s11A

To evaluate the three sources of error, three standards are employed. A 50 ohm reference
load effectively measures the directivity error ED since s11A = 0. A reference open and short
are used to evaluate ES and ET. Each of these reference terminations should be measured
over the same frequency range that will be employed to measure the device or circuit under
test, and the results stored. Once this is done, the device under test can be measured and its
actual reflection coefficient (s11A) determined.
For the two-port case a similar procedure is employed but the device or circuit under
test must be terminated in the system characteristic impedance. This should be as good as
Signal Transmission, Network Methods and Impedance Matching 201

the reference 50 ohm load used to determine the directivity value for the one-port situation.
For the two-port case, the most significant errors are source-mismatch, load-mismatch,
isolation and tracking. For this case, the actual s-parameters (sijA) of the device under test are
determined from the measured s-parameters (sijM) using:

A(1 + BESR ) − CDEL F


s11A = (3.224(a))
(1 + AESF )(1 + BESR ) − CDEL F EL R

C[1 + B( ESR − EL F )]
s21A = (3.224(b))
(1 + AESF )(1 + BESR ) − CDEL F EL R

D[1 + A( ESF − EL R )]
s12 A = (3.224(c))
(1 + AESF )(1 + BESR ) − CDEL F EL R

B(1 + AESF ) − CDEL R


s22 A = (3.224(d))
(1 + AESF )(1 + BESR ) − CDEL F EL R

where:

s11M − EDF
A= (3.225(a))
ERF

s22 M − EDR
B= (3.225(b))
ERR

s21M − EXF
C= (3.225(c))
ETF

s12 M − EXR
D= (3.225(d))
ETR

The error quantities in Equations (3.225) and (3.226) are identified in Table 3.3.

Table 3.3 Network analyser error terms

Symbol Error

EDF Forward directivity error


EDR Reverse directivity error
EXF Forward isolation error
EXR Reverse isolation error
ESF Forward source-mismatch error
ESR Reverse source-mismatch error
ELF Forward load-mismatch error
ELR Reverse load-mismatch error
ETF Forward transmission tracking error
ETR Reverse transmission tracking error
ERF Forward reflection tracking error
ERR Reverse reflection tracking error
202 Microwave Devices, Circuits and Subsystems for Communications Engineering

The above approach takes into account 12 error terms. As previously discussed, the error
terms are determined with 50 ohm, open and short reference terminations. The error value
should be determined over the same frequency range to be employed in the circuit testing.
The signal strength, cabling, connectors, etc. should be the same for the calibration and
circuit testing.
A variety of kits is available to perform the calibration procedure. These include kits for
connectors using 7 mm, type N, 3.5 mm, SMA, etc. Each kit includes a 50 ohm, short and
open termination. Reference adaptors are available to convert from one connector family to
another.
The effects of frequency and temperature drifts can be minimised by performing the
calibration process at regular intervals. Good RF cables and connectors are an integral part
of the system and good care should be taken of them. Errors such as those introduced by
connector repeatability and noise cannot be accounted for.

3.10.3 Transistor Mountings


The measurement of the s-parameters of a semiconductor device needs to accommodate two
situations, i.e. the measurement of a packaged or naked discrete device, and the measurement
of a discrete on-wafer device. These cases are considered in the following subsections.

1. S-parameter measurements of packaged devices


For a packaged device such as a bipolar transistor, a suitable device holder or jig is shown
in Figure 3.71. This consists of a brass body, usually gold-plated to minimise losses, which
supports two alumina substrates, each with a 50 ohm microstrip line. These lines are attached
to suitable high-frequency 50 ohm connectors (such as SMA) and via high-frequency 50
ohm cables to the network analyser ports. To eliminate the effects of any air gaps between
the substrate and the body of the jig, a film of conductive paste is applied to the underside of
the substrate. Sandwiched between the two halves of the jig is a brass spacer, which is used
to hold the device. The jig is designed so that different spacers can be utilised depending on
the size of the device package.

To Port 1
RF and DC To Port 2
RF and DC

SMA or other
suitable connector
Alumina substrate Brass gold plated body
with 50 Ω line Spacer to accomodate
packaged device

Figure 3.71 Microwave jig for packaged device measurement


Signal Transmission, Network Methods and Impedance Matching 203

Referene Plane Reference Plane

50 Ω Line DUT 50 Ω Line

Spacer width too large

50 Ω Line DUT 50 Ω Line

50 Ω Line DUT 50 Ω Line

Spacer width prevents


lateral placement errors

Figure 3.72 Spacer design

The spacer should be designed so that placement of the device on the jig is repeatable.
Placement differences between one device and another identical device will affect the
measured data, as shown in Figure 3.72.
The device itself is normally configured in common-emitter mode with the emitter terminal
connected to the body of the jig. The DC base current to the device is applied via the net-
work analyser RF port 1, while the collector-emitter voltage is applied via port 2. There is
therefore no need to have separate cables for biasing the device, and this improves the RF
performance of the jig. The leads of the device are not normally soldered to the 50 ohm lines
of the jig. A reasonable contact can be achieved with the use of a plastic clamp which does
not affect the accuracy of the RF measurements.
For a FET device the same jig can be used. In this case, the device would be configured
in common-source mode. Port 1 would be used to carry the gate-source DC bias, while port
2 would be used to apply the drain-source voltage.
Open-circuit calibration of the structure can be performed by leaving the 50 ohm jig lines
open. The through line calibration can be performed by bridging across the two 50 ohm lines
on the jig with a separate substrate containing a 50 ohm line. Alternatively, special substrates
can be designed for the jig to facilitate the calibration process.

2. S-parameter measurements of discrete naked devices


Due to their small size and lack of suitable ground connection points, the easiest way of
measuring such devices mechanically is to mount them onto an artificial package. The device
204 Microwave Devices, Circuits and Subsystems for Communications Engineering

SMA Ground Plane


Connector

Gold-plated
brass body
Ground Plane
50 Ω Line
Chip device bonded to
lines on mother board

Figure 3.73 Probing a chip device

Figure 3.74 Electrical equivalent circuit of mounting shown in Figure 3.73

connection pads can then be bonded to the artificial package or motherboard. From here,
SMA or other suitable connectors can be used to connect to the network analyser. A possible
scenario is shown in Figure 3.73.
The principal problem with such an approach is the de-embedding or calibration of the
system. In this respect the removal from the measured data of the parasitic effects introduced
by the bond wires and mounting of the device would present problems. This could be
overcome by developing an electrical model for the package, which can then be added to
the intrinsic model of the device. For the scheme shown in Figure 3.73, a possible model
would be as shown in Figure 3.74.
Assuming that the motherboard, 50 ohm lines, cables, connectors, etc. have been correctly
calibrated, the measured s-parameters would correspond to the model shown in Figure 3.74.
Notice that in the measurement of a packaged device, the same situation exists, i.e. the
s-parameters measured correspond to the intrinsic device and its package.
The estimation of the parasitic elements of the package model is difficult to do from a
measurement point of view. However, transmission line graphs can be employed to calculate
the series inductances and parallel capacitances. In essence, this approach compares the
package to a 50 ohm air line.
Signal Transmission, Network Methods and Impedance Matching 205

3. S-parameter measurements of on-wafer devices


The measurements of the s-parameters of a device on a wafer requires specialised probes in
order to maintain a 50 ohm environment to the device contact pads. Many foundry houses lay
out their devices as shown in Figure 3.75. The device is configured in common-emitter mode
and a co-planar design (ground–signal–ground) is employed for the device pads or contact
points. Each pad is typically 100 µm × 100 µm with a pitch between pads of 150 µm. To
make contact with the device, a probe such as the one shown in Figure 3.76 is employed.
The gold-plated brass body supports the connector and substrate. The substrate itself
uses a co-planar design for the lines, and the contact pads at the tip of the substrate are of
the same dimensions as those of the device on the wafer. To measure a transistor, two such
probes would be required. Commercially available probes have a frequency range in excess
of 40 GHz.
The probe would usually be attached to a micromanipulator to enable its position to be
finely controlled in the x, y or z direction. The device or the wafer sits on a prober as shown

Ground Pad
Ground Pad
(100 µm × 100 µm)

Signal Pad Signal Pad

Ground Pad Ground Pad

Figure 3.75 Device layout on a wafer

Ground line Substrate with 50 Ω


line and ground
50 Ω Signal connections
Line
SWA
Connector
Ground line Body of probe
Body
Plan View

Ground

Signal

Ground
Probe Tip

Figure 3.76 Co-planner microwave probe


206 Microwave Devices, Circuits and Subsystems for Communications Engineering

Microscope

Wafer
Micro-Manipulator

Wafer probe top plate

Vacuum Stage
xy stage

Figure 3.77 Wafer prober with micromanipulators and probes

in Figure 3.77. The wafer is held firmly on the stage by a vacuum pump and the stage is
moved up or down (to make or break contact with the probes) with air pressure. The stage
sits on an xy platform to enable different parts of the wafer to be probed without having to
move the probes or the wafer. The micromanipulators are held firmly on the top plate of the
wafer prober using a magnetic base. The positioning of the micromanipulator on the wafer
prober is not critical since the fine adjustment is done with the micromanipulator controls
and the xy stage of the wafer prober.
The calibration of the system is not as difficult as one might think. Many foundry houses
place calibration cells on the same wafer that contain the devices. These calibration cells
(open, through, 50 ohm load, short, etc.) employ the same co-planar design and dimensions
as the devices. Manufacturers of the microwave probes also provide calibration cells on a
wafer to aid the circuit designer. Some network analysers have built-in calibration procedures
which recognise such calibration cells.
Although the discussions in this section have been entirely based on the requirements for
measuring individual semiconductor devices, the need for careful attention to detail is also
important in the characterisation of circuits.

3.10.4 Calibration Approaches


In previous sections the calibration procedure highlighted requires four standards for the error
correction process. These standards are the short, open, load and through line (SOLT), which
enable the 12 terms of the error model to be evaluated. While this is the most common form
of calibrating system, it does have certain disadvantages that arise from the accuracy of
the standards themselves. Clearly the calibration standards must be accurately modelled and
consistent if the error correction process is to be accurate. In this respect the need for the
open and short standards presents a problem. For example, the open standard for on-wafer
Signal Transmission, Network Methods and Impedance Matching 207

measurements is usually realised by the probe tip being placed on an open circuit cell. For
this to work well, however, the capacitance of the probe tip must be accurately known. This
becomes more difficult to evaluate as the operating frequency increases. The same is true of
the short standard whose inductance must be accurately known for on-wafer measurements.
The 50 ohm load and through-line standards, on the other hand, are less difficult to model and
are more consistent. The SOLT method is best used in situations where the measurements
are carried out in a 50 ohm characteristic impedance environment.
The 12 terms of the error model can also be evaluated using the Through-Reflect-Line
(TRL) standard. As the name implies, this requires a through line and one or more transmission
lines of varying offset-lengths. This procedure avoids the problems with the open and short
standards needed for the SOLT method. The TRL approach therefore provides better accuracy
than the SOLT approach in on-wafer measurements where good quality line standards are
available. A disadvantage of the TRL approach is that, because the bandwidths of the lines
are limited, several lines may be required for broadband measurements. Some of these may
be physically long if low frequencies are to be included.
A third possibility is to use the Line-Reflect-Match (LRM) technique. In this case the
standards required are a transmission line for the through standard and broadband loads for
defining the impedance references. This procedure avoids the disadvantages of TRL and
SOLT since open, short or long line standards are not required and the standards that are
needed can be accurately fabricated and modelled.
The choice of calibration procedure is determined by the application, the availability
of standards and the availability of calibration models in the network analyser. Whichever
procedure is used, however, it should be remembered that the quality of the s-parameter
data for a device or circuit being characterised is determined by the accuracy with which
the system is calibrated.

3.11 Summary
Transmission lines, network methods and impedance matching techniques are all fundamental
tools for microwave circuit design. Transmission lines are principally used to connect other
components. They are also, however, passive devices in their own right, being the raw
material for couplers, power splitters and distributed filters. They can be implemented in a
variety of technologies including twisted and parallel pair, waveguide, coaxial cable, finline,
microstrip, slotline and other, more exotic, variations. Twisted pair, parallel pair, and coaxial
cable can support electric and magnetic fields that are purely transverse to the direction of
propagation and are therefore referred to as transverse electromagnetic (TEM) lines. Two-
conductor lines that have non-uniform dielectric constant in the plane orthogonal to the line
are quasi-TEM lines. TEM and quasi-TEM transmission lines are generally easy to analyse
because they allow distributed circuit theory to be applied. The analysis of other non-TEM
technologies is generally more difficult since it requires the application of field theory.
Microstrip is the most important transmission line technology in the context of circuit
design. The microstrip design problem is typically that of establishing the line geometry
(width and length) given the required operating frequency, characteristic impedance and
parameters of the substrate on which the line is to be implemented (i.e. relative permittivity
and thickness). This is usually achieved using semi-empirical design formula (or curves)
often implemented within a CAD software package.
208 Microwave Devices, Circuits and Subsystems for Communications Engineering

Dispersion in microstrip cannot be avoided completely but can be predicted using field
theory or approximated using semi-empirical models. Other microstrip limitations such as
surface waves and transverse resonance may also need to be considered in some designs.
A discontinuity occurs whenever the geometry of a microstrip changes. The most com-
mon types of discontinuity are the foreshortened open end, microstrip via, the right-angled
bend and the T-junction. The open end results in fringing capacitance that is often modelled
by an additional (hypothetical) section of line. The microstrip via is a plated through hole
that ‘short-circuits’ the microstrip conductor to the ground plane. The discontinuity of a
right-angled bend can be mitigated by appropriately mitring (chamfering) the outside corner
of the bend and the optimum effective widths of the main and branch lines comprising a
T-junction can be found using semi-empirical models.
Microstrip can be used to implement a range of low-Q filters. Designs are typically
realised by transforming the lumped elements (capacitances and inductances) of a conven-
tional filter into equivalent segments of parallel and series connected microstrips.
The electromagnetic coupling that occurs between closely spaced microstrip lines can be
used to realise a variety of devices including filters and directional couplers. The design of
these devices typically relies on the theory of even and odd modes – the former referring to
the component of electric field that has the same polarity on each microstrip and the latter
referring to the component of field that has opposite polarity.
A general two-port device or circuit can be described from a system’s point of view
by any of several sets of network parameters. These include impedance- or z-parameters,
admittance- or y-parameters, hybrid- or h-parameters, ABCD parameters and scattering- or
s-parameters. The latter are by far the most important at microwave frequencies although, in
principle, all sets contain identical information and transformation between sets is possible
using simple matrices. The advantage of s-parameters relates to the practical difficulty of
engineering good open- and short-circuit loads (precisely) at the terminals of the device or
circuit under test, such loads being required for all except s-parameters. Their interpretation
as travelling wave reflection and transmission coefficients is also particularly appropriate
and intuitive at microwave frequencies.
Impedance matching refers to the design of circuits and subsystems such that their output
and input impedances over a required frequency band realise some overall performance
criteria such as maximum power transfer, minimum return loss or minimum noise figure
(see Chapter 4). The Smith chart provides a graphical representation of impedance and is
commonly used as a design tool for impedance matching problems.
Network analysers are used to measure the s-parameters of a circuit or device automatically.
Accurate measurements require prior calibration of the network analyser and associated
cables, connectors and mounts.

References
[1] S.Y. Liao, Microwave Circuit Analysis and Amplifier Design, Prentice Hall Inc., Englewood Cliffs, NJ, 1987.
[2] F.A. Benson and T.M. Benson, Fields Waves and Transmission Lines, Chapman & Hall, London, 1991.
[3] C. Bowick, RF Circuit Design, H.W. Sams & Co., Indianapolis, 1982.
[4] G.D. Vendelin, A.M. Pavio and U.L. Rohde, Microwave Circuit Design Using Linear and Nonlinear Techniques,
Wiley & Sons, New York, 1992.
[5] V.F. Fusco, Microstrip Circuits, Prentice-Hall, Englewood Cliffs, NJ, 1989.
[6] T.C. Edwards and H.B. Steer, Foundations of Interconnect and Microstrip Design, 3rd edition, John Wiley &
Sons, Chichester, 2000.
Amplifier Design 209

4
Amplifier Design
N. J. McEwan and D. Dernikas

4.1 Introduction
Everybody knows that the job of an amplifier is in some way to increase power, and would
have no hesitation in defining gain as output power over input power. To make these ideas
precise enough to use in amplifier design work, we have to think more clearly and define all
the power quantities carefully.
Section 4.2 will first give the basic definitions of gain, and derive expressions for gain,
working in terms of s-parameters. There will then be a comment on the mathematical origin
of the two types of expressions that give rise to graphical representations in terms of circles.
In the final part of this section, the idea of gain circles will be briefly introduced.
We then digress in Section 4.3 to look at some basic ideas of stability – a major con-
sideration for any amplifier design is to avoid instability or oscillation. After that, we
return to discussing the appearance of gain circles, and some further concepts relating to
power gain. Section 4.4 looks into techniques for developing broadband amplifiers, while
Section 4.5 concentrates on what is typically the first element in any amplifier – the low
noise amplifier.
The limitations and features of implementing amplifiers in practice are discussed in
Section 4.6, as at microwave frequencies the capacitance, inductance and resistance of the
packages holding the devices can have very significant effects. As we increase our operating
frequency, alternative circuit layout techniques can be used in place of discrete inductors or
capacitors, and some of these are discussed. In dealing with this we see that even relatively
simple amplifier circuits are described by a large number of variables. The current method
of handling this amount of data and achieve optimum designs is to use a CAD package, and
some of the more popular packages are described in Section 4.7.

4.2 Amplifier Gain Definitions


Consider first the situation shown in Figure 4.1, where a linear two-port network is connected
to a given load and a given generator. Remember that we are working in the frequency

Microwave Devices, Circuits and Subsystems for Communications Engineering Edited by I. A. Glover, S. R. Pennock
and P. R. Shepherd
© 2005 John Wiley & Sons, Ltd.
210 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 4.1 Layout of two-port with generator and load

domain. We are considering what is happening at a single frequency; all the quantities we
are using are functions of frequency, but we do not spell this out.
It should be remembered that, while an impedance is an intrinsic property of a one-port
device on its own, its reflection coefficient relates to its interaction with its environment, and
is only defined when the reference impedance of that environment has been specified. The
same comments apply, for a multi-port device, to its impedance matrix and its s-parameter
set. All the reflection coefficients and s-parameters in Figure 4.1 are assumed to have been
expressed relative to a specified reference impedance, in practice, probably 50 Ω.
The two-port could just be a single transistor, maybe a complete amplifier, or one of
many other devices. In a typical amplifier problem we have an active element as a central
block, and this is most commonly a single transistor, possibly with some feedback com-
ponents. On each side of the device we have a pair of impedance matching networks
designed to couple the device to a given generator and load. Usually it will be convenient
to take the reference planes in Figure 4.1 as lying on each side of the active device. We can
then choose the generator and load reflection coefficients, or equivalently their impedances,
as we please, by varying the matching networks, but the s-parameters of the two port will
be fixed.
We could, as an alternative, move the reference planes out to the final generator and load
positions, and include the matching networks inside the central two-port. In this case the
source and load Γs would be fixed and the s-parameters of the two-port would change if the
matching networks were varied.
However, it is probably simplest to use the inner reference planes and visualise the effect
of changing the load and generator Γs. Fortunately the matching networks are mostly chosen
to be nominally lossless, meaning that in practice their losses will usually be negligible,
at least as a first approximation. In this case the power gains defined at the inner reference
planes will then automatically be the same as those defined at the outer ones.
Amplifier Design 211

Referring to Figure 4.1, at the input plane, we can define two power quantities:

1. The power available from the generator, PA,G.


2. The actual input power Pin accepted by the two-port from the generator.

Note that Pin is the net power flow from left to right at the input reference plane, and it will
be represented as the difference in powers carried by the incident (right travelling) wave and
reflected (left travelling) waves at that plane.
It is very important to be clear what parameters of the system the various power quantities
depend on. For example, note that PA,G is a property of the generator alone, while Pin depends
on its interaction with the two-port. In the general case where the two-port is not unilateral,
i.e. has some coupling from its output to its input, varying the load impedance on the output
port can change the apparent impedance or reflection coefficient at the input port, and hence
can affect Pin. So Pin, in the general non-unilateral case, depends on all four s-parameters of
the two-port, on ΓL, and on both PA,G and ΓG.
Now looking at the output port, we can clearly define a similar pair of quantities:

1. The actual power PL accepted by the load from the two-port.


2. The available power PA,out at the output port of the two-port.

The physical meaning of PL is clear, but what does it depend on? The output port of the
two-port becomes a well-defined generator if, and only if, its output impedance and available
power are defined. These in turn are well defined if and only if generator properties ΓG and
PA,G are well defined. The load power PL also obviously depends on how much power is
being reflected back from the load, and hence on ΓL.
In fact, PL depends on all the parameters of the system – all four s-parameters of the two-
port, ΓG, PA,G and ΓL. However, the quantity PA,out is, by definition, the quantity PL when it is
optimised with respect to ΓL while keeping all the other parameters fixed. Thus PA,out is a
function of all parameters except ΓL. These points will become clearer when the flow graph
analysis of the system is undertaken.

4.2.1 The Transducer Gain


It turns out that the most fundamental definition of power gain is the ratio of actual load
power to available power from the generator. This is called the transducer gain:

PL
Transducer gain Gt = (4.1)
PA,G

Why is this is the most fundamental definition? To do something practically useful, an


amplifier must deliver power into a real, specified load, which might be some signal process-
ing device such as a demodulator, another amplifier, an antenna, a transmission line for
communicating signals to some distance point, etc. It must do this when fed from an actual
specified signal source, and the measure of its practical usefulness is that the actual power
we get into our load is greater than the maximum power we could have obtained from
the specified source. The practical usefulness does not, however, depend on whether the
amplifier is actually accepting all the power available from the source.
212 Microwave Devices, Circuits and Subsystems for Communications Engineering

You should by now be convinced that the transducer gain depends on all the parameters
of the system: all four s-parameters of the two-port, ΓG, and ΓL. We can write:

GT = GT (ΓG, [S], ΓL) (4.2)

where [S] denotes the two-port’s s-parameter matrix, to emphasise this functional dependence.
This again suggests that GT is the most fundamental definition, as it takes all the relevant
parameters of the system into account.

4.2.2 The Available Power Gain


This quantity, as its name suggests, is simply defined as the available power at the two-port
output over the available power at the input

PA,out
Available power gain GA = (4.3)
PA,G

To interpret this, imagine that everything is fixed except the load, which we vary to make
PL as large as possible. GA is then this load power divided by the available generator power.
GA is thus the same as GT optimised with respect to ΓL while keeping everything else fixed.
Because it has been optimised with respect to ΓL, it is no longer a function of it; thus, we
can write:

GA = GA(ΓG, [S]) (4.4)

Why might we wish to use this alternative definition? There are three possible reasons:

1. It gives us a simpler expression to work with, by allowing us to ‘shelve’ one of the


dependencies. It is easy to find an explicit expression for GA and it tells us the transducer
gain that will be achieved on the assumption that the output is matched for optimum power
transfer, but without needing to worry about the actual value of ΓL that is required to do
it. The problem of actually obtaining the output match is thus put aside for the time being.
When we use the expression for GA to examine its variations with ΓG, it is as though
the output matching network is constantly being adjusted to maintain an output match by
tracking the variations in the output impedance of the two-port caused by the variations
of the generator impedance.
2. It can be used in cascading. If we cascade a pair of two-ports, labelled 1, 2, 3 . . . N
with 1 at the input, the overall transducer gain can be written as GA,1 · Gt,2. This is not quite
as simple as it looks, because the value of the second gain term depends on the output
reflection coefficient of the first stage – which in turn depends on the generator impedance
at the input to the first stage. In system design, however, it is usual to perform a straight-
forward multiplication which will be approximately correct provided the cascaded ports
can be taken as reasonably well matched to the reference impedance level.
3. As will be discussed elsewhere, GA appears in several expressions relating to noise
theory, especially the formula for the overall noise figure of cascaded networks, and it is
the appropriate quantity to use when choosing ΓG to optimise the trade-off between gain
and noise figure.
Amplifier Design 213

4.2.3 The Operating Power Gain


This is the most obvious definition and it is simply the actual output power over the actual
input power:

PL
Operating power gain GP = (4.5)
Pin

The reasons for using this definition are just like the reasons (1), (2) given above for using
the available gain definition, but the reason (3) relating to noise does not apply:

1. Gp allows us to evaluate GT and its dependence on ΓL on the assumption that an


input match is always maintained, and without worrying about the value of ΓG actually
needed.
2. Gp can be used in other ways of writing formulae for cascaded power gain, which is left
as a self-assessment exercise, see Self-assessment Problem 4.1.

Self-assessment Problem
4.1 For two cascaded two-ports:
(a) Show that the overall available power gain is the product of the individual
available power gains. Does the same hold for operating power gains?
(b) Show that the overall transducer gain, which we earlier wrote as GA,1 · Gt,2,
could also be written as Gt,1 · GP,2.

4.2.4 Is There a Fourth Definition?

You may have noticed that there is a fourth combination of powers we could use to try to
define a power gain, namely, PA,out /Pin. This is well defined if we imagine the two-port with
a fixed generator, and the output load being adjusted to receive the maximum power from it.
It turns out not to be a particularly useful definition.
We might, however, re-interpret this definition by forgetting about the generator and just
adjusting the load to maximise the ratio of PL /Pin. In that case we would obtain Gp optimised
with respect to ΓL , or GT simultaneously optimised with respect to both ΓG and ΓL. Either way,
the result would depend on [S] only, i.e. it would be an intrinsic property of the two-port.
This does prove to be an important quantity and it is discussed further after considering the
issue of stability.

4.2.5 The Maximum Power Transfer Theorem


Before proceeding to derive expressions for the various power gains, we shall first do two
preliminary derivations. The first will be to revisit the well-known maximum power transfer
theorem, but examining it in terms of s-parameters.
214 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 4.2 Generator connected to a load

Consider a generator and load connected as in Figure 4.2. As normally stated, the optimum
or matched load which can extract the maximum power (or ‘available’ power) from the
generator must be chosen to have:

ZL = Z *G (4.6)

Possibly you only learnt this theorem in the simple case where both the source and load
impedances are real, i.e. purely resistive. If so, you can easily see what the complex statement
is saying. Let us write ZG = RG + jXG, ZL = RL + jXL. Now suppose that RL has been fixed.
Then, if the total reactance in the circuit is non-zero, it can only reduce the current flowing
and hence reduce the power delivered to the load. Hence, part of the optimisation must be
to choose XL = −XG. This is saying that the load reactance will be chosen to cancel out the
generator internal reactance, making the total reactance zero. Then, since we know that the
final optimisation must have this value of load reactance, we can just fix it at that value, and
the problem then reduces exactly to its simple form involving only resistances. We then
know from the elementary derivation (which need not be repeated, as it only needs simple
calculus to find the maximum) that the optimum load resistance is the same as the source
resistance. It has thus been shown that the optimum load is ZL = RG − jXG = Z *. G
The problem will now be described in s-parameter terms. While impedances in a system
have absolute values, the s-parameters only acquire definite values in relation to some specified
reference impedance level. Thus a reflection coefficient Γ, which is the single s-parameter of
a one-port network, is related to the associated impedance by the equation:

Z − Z0
Γ= (4.7)
Z + Z0
Amplifier Design 215

Figure 4.3 Flow graph for generator connected to load

where Z0 is the reference impedance. The convention will be adopted here that the reference
impedance Z0 is always assumed to be real, and of course in practice usually 50 Ω. This
convention makes it unnecessary to distinguish between power waves and travelling waves
and, in any case, we know that transmission lines at high frequencies have a real Z0 to a very
good approximation.
Now imagine that the generator is connected, not to the actual load, but to an infinite
transmission line whose characteristic impedance is Z0. A parameter a0 will be defined as the
complex wave amplitude, using the generator terminals as the reference plane, that the
generator would launch into that line. There is no reflection on the line because it is infinite,
but the generator has a non-zero reflection coefficient ΓG looking back into its terminals
from the line, because its impedance in general is not equal to Z0.
If the generator is now re-connected to the actual load, a flow graph can be drawn as in
Figure 4.3. The two coefficients a, b are the coefficients of forward and reverse waves. There
is no need to have any transmission line in the system at all. If it is helpful you imagine a
very short section of line interposed between the load and generator terminals, and visualise a
and b as specifying the amplitudes and phases of the forward and reverse waves on that line.
Mason’s rule (Appendix II) applied to this flow graph gives:

a0
a= , b = ΓL a (4.8)
1 − ΓG ΓL

The load power is now the forward wave power minus the reverse wave power:

Load power = | a | 2 − | b | 2 = | a | 2 · (1 − | ΓL | 2 ) (4.9)

Therefore, using the expressions for a and b:

| a0 |2 ⋅ (1 − | ΓL | 2 )
Load power = (4.10)
|1 − ΓG ΓL | 2

Now we know that the optimum load is ZL = Z *, G and if we insert this value into the Γ − Z
relationship, equation 4.7, we shall find that the corresponding optimum load reflection is
also the conjugate of the generator reflection coefficient:

Optimum (matched) ΓL = Γ *G
216 Microwave Devices, Circuits and Subsystems for Communications Engineering

NB The assumption that Z0 is real is essential to this last step. Then if this value of ΓL is
inserted into the expression for load power, we shall have an expression for the available
load power:

| a0 | 2
Available power = (4.11)
1 − | ΓG | 2

This result will be used later in finding expressions for power gain.

Self-assessment Problem
4.2 (a) Combine Equations (4.10) and (4.11) to obtain an expression, in terms of ΓG,
ΓL for the fraction of the available power that is delivered to the load.
(b) If a 50 Ω source is connected to a 100 Ω load (both being resistive), use your
equation, assuming 50 Ω reference impedance, to show that the load power is
0.51 dB less than the available power.
(c) Verify that the same result is obtained if the reference impedance is taken as
25 Ω.
(d) In Equation (4.10), suppose that | ΓL | is fixed but you are allowed to change its
argument (phase angle). How would you choose its phase to obtain maximum
power transfer? (Think about a phasor diagram representing the denominator
of the equation.) Could you use this result to finish proving the maximum
power theorem?

It is very important to be aware of the limitations under which the maximum power theorem
was derived. It was assumed that everything stays absolutely linear no matter what load
impedance is chosen, the generator’s internal excitation stays fixed, and the load is being
varied to extract the maximum possible output power for this fixed excitation.
This optimisation can be very different from the one which applies when optimising
the load in a power amplifier. The object there is usually to maximise the saturated load
power, the power at which non-linearity sets in, but the input excitation level can also be
varied at will in achieving the best saturated power – provided the power gain does not
become unacceptably low. Slight variations of this optimisation are to optimise the DC to
RF conversion efficiency, or power added efficiency, both near saturation. In all these cases
the two main conditions for deriving the simple theorem fail. Hence, in power amplifiers
designed to work near saturation, the load impedance may be far from the conjugate of the
output impedance of the active device.

4.2.6 Effect of Load on Input Impedance


Before proceeding to derive expressions for the various power gains, we shall do one final
simpler derivation, which is connected with the gain expressions and is used later on in the
theory of stability. If we have a two-port device or network, which is not unilateral, there
is some coupling from the output port back to the input. The effect is to make the input
Amplifier Design 217

Figure 4.4 Flow graph for considering effect of load on input reflection coefficient

impedance dependent on the load impedance, and the output impedance dependent on
the generator impedance. Single transistors always show this effect, which is crucial when
considering stability.
It is a good exercise in flow graph analysis to derive the expression which describes
the effect of a varying load impedance, connected at the output side of a two-port, on
the apparent impedance seen looking into the input port. We work, however, in terms of
reflection coefficients rather than impedances, as shown in Figure 4.4. When terminated
by a specified load, the two-port acquires a definite value for the apparent reflection
coefficient seen at its input port. (Taken together with its load, it becomes in effect a
one-port.)
N.B. The notation Γin will be used throughout for the input reflection coefficient of a
two-port, when made definite by specifying the load. Note, however, that some books use
S′11 for the same quantity. Obviously the analogous notation Γout (or S′22 ) will be used
when considering how the output reflection of the two-port is modified by the generator
impedance at its input.
In this simple graph we can cut out a step or two from the formal Mason’s rule recipe.
There is only one first-order loop in the system. There are clearly just two paths from a1
to b1.
If we just had the second path via S21, ΓL and S12, Mason’s rule would give the transfer
function of this path as S21ΓL S21/(1 – S22ΓL ). In fact, we have a second path via S11 but this
has no associated loops and is obviously unaffected by the loop on the other path. Hence the
total transfer function is just the sum of the two individual ones and we get:

S21ΓL S12 S − ∆ΓL


Γin = S11 + = 11 (4.12)
1 − S22 ΓL 1 − S22 ΓL

The first expression here is the one obtained by summing the two transfer functions, and the
last one is just a neater but exactly equivalent way of writing it – the symbol ∆ is a standard
notation for the determinant of the s-parameter matrix, which is S11S22 – S21S12.
It should be obvious, by symmetry, that we could get the corresponding expression for
Γout by just interchanging suffices ‘1’ and ‘2’ in all the expressions, and of course replacing
ΓL by ΓG. (As far as the maths is concerned, it doesn’t matter which way round the ports of
the two-port are labelled!)
218 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 4.5 Flow graph for two-port with generator and load

4.2.7 The Expression for Transducer Gain


To derive the expression for transducer gain we now need a slightly more elaborate flow
graph, Figure 4.5, which is like Figure 4.4 except that the generator has been added. This
graph describes the physical set-up shown in Figure 4.1. The relationships between the
generator parameters PA,G, a0 and ΓG have already been discussed. At one frequency, just
two parameters – ΓG, and either PA,G or a0 – are sufficient to characterise the generator,
provided it remains linear. The main step in finding the transducer gain is to write down the
transfer function b2 /a0. We also need to know a2, but this is just ΓL times b2.
There are three first-order loops in this graph, which are the two obvious end loops, and
the outermost loop formed by the arrows S21, ΓL, S12, and ΓG. A second-order loop is defined
as the product of any two non-touching first-order loops – in this case there is just one, the
product of the two simple loops at each end. All of these loops touch the single path from a0
to b2 which is formed by the arrows 1 and S21.
Since all the loops touch the path, Mason’s rule says that the transfer function is the path
product divided by (1 – sum of all first order loops + sum of all second order loops.)
This reads:

b2 S21
= (4.13)
a0 1 − S11ΓG − S22 ΓL − S21ΓL S12 ΓG + S11ΓG S22 ΓL

On inspection, a slightly neater way of writing this can be seen:

b2 S21
= (4.14)
a0 (1 − S11ΓG )(1 − S22 ΓL ) − S21ΓL S12 ΓG

Now the load power is:

PL = | b2 | 2 − | a2 | 2 = | b2 | 2 · (1 − | ΓL | 2 ) (4.15)

so that Equation (4.12) will now give:

PL | S21 | 2 (1 − | ΓL | 2 )
= (4.16)
| a0 | 2
| (1 − S11ΓG )(1 − S22 ΓL ) − S21ΓL S12 ΓG | 2
Amplifier Design 219

Table 4.1 Transducer gain expressions

Three equivalent expressions:


(1 − | ΓG | 2 ) | S21 | 2 (1 − | ΓL | 2 )
GT =
| (1 − S11ΓG ) (1 − S22 ΓL ) − S21ΓL S12 ΓG | 2
(1 − | ΓG | 2 ) (1 − | ΓL | 2 )
GT = ⋅ | S21 | 2

| 1 − Γin ΓG | 2 | 1 − S22 ΓL | 2
(1 − | ΓG | 2 ) (1 − | ΓL | 2 )
GT = ⋅ | S21 | 2 ⋅
| 1 − S11ΓG | 2
| 1 − Γout ΓL | 2

Unilateral approximation:
(1 − | ΓG | 2 ) (1 − | ΓL | 2 )
GTU = × | S21 | 2 ×
| 1 − S11ΓG | 2
| 1 − S22 ΓL | 2
= Product of 3 separable ‘gain’ factors:
One dependent on One intrinsic One dependent on
input match output match

Note: S ′11, S ′22 as alternative notation for Γin, Γour in some texts.

Now refer back to Equation (4.11), and recall the definition of transducer gain, and it should
be obvious that:

(1 − | ΓG | 2 ) | S21 | 2 (1 − | ΓL | 2 )
GT = (4.17)
| (1 − S11ΓG )(1 − S22 ΓL ) − S21ΓL S12 ΓG | 2

Now refer to Table 4.1 which contains the above important expression for transducer gain,
and two further ways of writing it. These are exactly equivalent, as can be shown after some
algebra, using Equation (4.12) or the corresponding expression for Γout.
Rather than doing the algebra, there are two ways of re-drawing the flow graph in
Figure 4.5, which makes the equivalence of the three expressions obvious, see Figure 4.6.
In this version of the flow graph, we have removed the feedback path and represented
instead its effect on the input reflection coefficient by substituting Γin for S11. The four paths
representing the two-port are no longer generally equivalent to [S], because Γin depends on

Figure 4.6 Modified flow graph


220 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 4.7 Second modified flow graph

ΓL, but they are equivalent for the particular ΓL in question. Looking into the input port, the
real system and the one described by Figure 4.6 are indistinguishable. This implies that the
input a1, b1 variables must be the same. However, if a1 is the same, then b2 and a2 must also
be −a1 as the only input to the right-hand section of the diagram; the feedback path only
leads away from that section and its removal cannot affect the variables on the output side as
long as the input variable a1 stays the same. It should by now be obvious that the second form
of the expression for GT could be derived immediately from the modified flow graph.
Another way of re-drawing the flow graph is shown in Figure 4.7. The feedback loop
causing Γout to be affected by ΓG has been separated out. Knowing that the system is linear,
we could superpose the nodes a 1″, b 1″ on top of a′1, b′1 and we would have the original flow
graph of Figure 4.5 again, with the relations: a1 = a′1 + a 1″, b1 = b′1 + b 1″. It would then be
possible to reduce the dashed section of Figure 4.7 to a single path with a transfer function
Γout, so that the graph could be drawn yet again as Figure 4.8. Although the a and b variables
on the left-hand side are now no longer the same as in Figure 4.5, the output variables a2
and b2 are the same, and so are the output power and the available generator power. Thus,
Figure 4.8 could be used in a simple derivation of the third form of the GT expression.
The final expression in Table 4.1 is the form that the GT expression would take if the
device could be taken as unilateral, S12 = 0. The validity of this unilateral approximation
depends both on the device and the source and load impedances, and often works well for
transistors at high frequencies and/or when designing for gains considerably below the
maximum that the transistor could give at the frequency in question.
In the unilateral approximation the GT expression breaks up simply into three factors:
the first dependent only on the input match, the coupling factor | S21 | 2 which is intrinsic to
the two-port, and the third dependent on the output match. The first and third factors are
Amplifier Design 221

Figure 4.8 Final modification to flow graph

sometimes referred to as the input and output matching ‘gains’. They can be larger than unity,
allowing GT to exceed | S21 | 2.
It is important to keep in mind the possibility of using (with caution) the unilateral approx-
imation, as it simplifies the design problem by separating the input and output matching
processes. Even if the approximation is not too good, it may help us to make a rough design
which a CAD package, working with the accurate expressions, could be allowed to optimise.

4.2.8 The Origin of Circle Mappings


In RF and microwave engineering, important quantities frequently appear in the form of
circle diagrams. This happens in two general ways, and we pause here for a short comment
on the mathematics involved.
The first general type of circle diagram arises from a bilinear mapping:

Az + B
ζ = (4.18)
Cz + D

In this equation A, B, C, D are complex constants, and z should be thought of as the variable.
For any value of z in the complex plane, a complex value of ζ is generated. The equation can
thus be pictured as mapping the z plane into the ζ plane.
The two main things to remember about the bilinear mapping are:

1. The bilinear transformation always maps circles in one plane into circles in the other.
(The term circles includes straight lines as a special case i.e. circles of infinite radius.)
2. The area inside a given circle in the z plane may be mapped into either the interior of the
corresponding ζ plane circle, or the entire region exterior to it.

Thus, if we draw a circle in the z plane which encloses the point z = −C/D, ζ will go off to
infinity at this point, so we know that the interior of the z circle is mapped into the exterior
of the ζ circle. Any z circle not enclosing this critical point would have to be mapped
interior to interior, because ζ would have to stay finite at all interior points of the z circle.
A familiar example is the Γ ↔ Z relation (Equation 4.7), leading to the Smith chart in which
straight lines of constant resistance or reactance in the impedance plane correspond to the
circles in the reflection coefficient (Γ) plane.
222 Microwave Devices, Circuits and Subsystems for Communications Engineering

Another important application is the stability circles to be discussed later. Many situations
in RF engineering give rise to bilinear expressions; error correction in a network analyser is
another example.
The second type of circle diagram arises in a different way. Let z be a complex number,
and write x and y for its real and imaginary parts: z = x + jy. Now consider an expression of
the form:

Ax 2 + Ay 2 + Bx + Cy + D
X= (4.19)
Ex 2 + Ey 2 + Fx + Gy + H

In this expression, all the constants are real, and so is the result X. The expression can be
visualised as associating a real number with any point in the complex z plane.
What are the loci in the z plane along which X takes a constant value? Again, the answer
is a circle. This is easy to prove, as is discussed in Appendix I, both for these and for the
bilinear mappings.
The distinguishing feature of the above expression which gives rise to the circle feature is
that the coefficients of x 2 and y 2 in the numerator are equal, i.e. both A, and likewise their
coefficients are both E in the denominator. If x 2 and y 2 had unequal coefficients on either
top or bottom, the loci of constant X would become ellipses.

4.2.9 Gain Circles


The gain circles are simply the loci, in either the ΓG or ΓL planes, of any one of the various
power gain quantities defined above. It can easily be seen that the various expressions
always have the form of the second type of expression discussed in the previous section, so
the locus of constant gain is always a circle.
The following are the types of gain circle we might consider plotting:

1. Circles of constant GT in the ΓG plane, for a fixed value of ΓL.


2. Circles of constant GT in the ΓL plane, for a fixed value of ΓG.
3. Circles of constant GA in the ΓG plane.
4. Circles of constant GP in the ΓL plane.
5. Circles of constant value of input matching gain, in the ΓG plane.
6. Circles of constant value of output matching gain, in the ΓL plane.

The input and output matching gains are the factors discussed above in connection with
the unilateral approximation. For a strictly unilateral device, i.e. S12 = 0, all the ΓG plane
circles mentioned will reduce to just one family, and likewise all the ΓL circles.
The gain circles are readily plotted by CAD packages and are a useful design aid. It is
useful to have a group of circles plotted at a selection of frequencies spanning the range we
are trying to design for. The trajectory of Γ, as frequency varies, generated by a projected
matching circuit, can be plotted against the background of the circles, and used to visualise
how the gain will vary. It may be possible to guess a matching circuit topology which will
generate a locus that tracks across the gain circles in a way that gives a flat gain or a desired
rate of gain tapering. An example will be discussed later. The general appearance of gain
circles will be discussed further after considering the question of stability.
Amplifier Design 223

Self-assessment Problem
4.3 (a) GT will become equal to GA when ΓL is set equal to Γ *out. Use the third form of
the expression for GT to show that:

(1 − | ΓG | 2 ) 1
GA = ⋅ | S21 | 2 ⋅
|1 − S11ΓG | 2 (1 − | Γout | 2 )
(b) We showed earlier that:

S22 − ∆ΓG
Γout = 1.
1 − S11ΓG
Use this to prove that:
1
GA = (1 − | ΓG | 2 ) ⋅ | S21 | 2 ⋅
|1 − S11ΓG | 2 − | S22 − ∆ΓG | 2

(c) If a and b are any two complex numbers, the modulus squared of their sum
can be obtained as:

| a + b | 2 = (a + b) · (a + b)* = a · a* + a · b* + b · a* + b · b*
= a · a* + b · b* + a · b* + (a · b*)* = | a | 2 + | b | 2 + 2Re(ab*)

Try using this result to expand one of the numerator terms in the expression just
obtained for GA, for example, the term | S22 − ∆ΓG | 2. Writing ΓG = x + jy, you should
find that the only terms of quadratic order occur in | ∆ | 2 · |ΓG | 2 which is | ∆ | 2 · (x 2 + y 2).
This, of course, is in the form required to generate circles as mentioned before.
You should be able to convince yourself that the same thing happens in the other
terms in this expression, and indeed all the various expressions for gain quantities, and
hence that the loci of constant gain are always circles.

4.3 Stability
Unintentionally creating an oscillator is the most notorious pitfall of amplifier design.
The resulting circuit is usually quite useless for its intended purpose, and effort put into
an elaborately optimised design, e.g. for low noise or flat gain, will have been wasted. An
indifferently performing but stable amplifier would be more useful.
Stability is therefore the first and most vital issue that must be addressed in amplifier
design. There are several reasons why unintended oscillation is an easy trap to fall into:

1. We are tempted to do our design work, e.g. of impedance matching networks, only for
the intended operating frequency band of our circuit but oscillation might occur at any
frequency, often one far removed from the design frequencies.
2. Virtually all single transistors are potentially unstable at some frequencies, which are
often outside the design frequency band.
224 Microwave Devices, Circuits and Subsystems for Communications Engineering

3. The source and load impedances are a critical factor in producing oscillation, but are
never wholly predictable. Oscillation is most likely with highly reactive source or load
impedances, and these are very likely to occur outside the operating band.

The general problem of the predictability of the impedance environment is very important.
The external impedances are least predictable outside the design operating band, especially
in the ‘building block’ approach to system design where cables of arbitrary length (producing
unpredictable impedance transformation) may be used to interconnect the functional units.
Outside their specified operating band, the external devices will often have nearly unity
reflection coefficients, especially components such as antennas and filters.
Even within the operating band, the external impedances may be uncertain. We could, for
example, be designing an amplifier to feed into a nominally 50 Ω load, but then connect it to
another amplifier made by someone else which is designed to be fed from 50 Ω but whose
input does not itself present a close approximation to 50 Ω – some mismatch may have been
designed in to produce flatter gain or better noise figure.
When designing amplifiers we should therefore try to have some feel for what regions of
the Smith chart our source and load Γ’s could possibly lie in, and especially in the frequency
ranges of potential instability. Even when designing for an ideal 50 Ω external environ-
ment, this must still be subject to some tolerance. The following sections will consider how
oscillation can occur, and how it can be avoided.

4.3.1 Oscillation Conditions


A criterion is now needed to establish the conditions under which a circuit will oscillate, both
for the purpose of avoiding oscillation in amplifiers, and to deliberately design oscillators
when we need to.
When designing amplifiers it is mostly sufficient to consider a ‘one-port’ oscillation
condition, and the following discussion will be limited to this. First, consider a one-port
network (one pair of terminals) at which we know the impedance Z( f ) = R( f ) + jX( f ) as a
function of frequency (Figure 4.9(a)). Under what conditions will the network oscillate if the
terminals are short-circuited?

Figure 4.9 Tests for oscillation: definitions of quantities


Amplifier Design 225

When the network is completely general, it is quite difficult to state a rigorous condition
for oscillation to occur. However, we can limit our discussion to simple networks composed
of just one transistor, with common emitter or source, and associated passive matching
networks. For these simple cases a criterion which will usually work is that the circuit will
be oscillating with short-circuited terminals if the plot in the complex plane of Z( f ), as f runs
from −∞ to ∞, crosses the negative real axis at a finite frequency with the reactance increas-
ing with frequency at the crossing point. In other words there is a finite frequency where
R < 0, X = 0, and dX/df > 0. At this frequency we would also find conductance G < 0,
susceptance B = 0 and dB/df < 0. (Using admittance Y( f ) = 1/Z( f ) = G( f ) + jB( f ).)
The condition described, where the impedance plot crosses the negative real axis, is
the condition that the circuit when first switched on will exhibit exponentially growing
oscillations, and of course in a real circuit the oscillation amplitude will always be limited
by the onset of non-linearities after a very short time. You might find it somewhat easier to
think of the special case where the impedance plot passes exactly through the origin at some
finite frequency – in this case, the circuit is in the critical condition where oscillations at
that frequency will, if started, neither grow or decay in amplitude. Physically we are saying
that the total circuit impedance at that frequency is zero, and hence that a current at that
frequency can exist without any voltage to drive it.
Now consider the situation shown in Figure 4.9(b). Block 2 represents a transistor, or
other active device. The terminals k, h could be the emitter and base, or source and gate, and
there will be a passive load network connected at the output terminals (usually emitter and
collector, or source and drain, i.e. using the common emitter or common source connection).
The input impedance Zin( f ) will be determined once the s-parameters of the device are
known, and the load network has been specified. Block 1 with impedance ZG( f ) will be a
passive network, generally the generator with associated matching/filtering circuitry. (The
discussion will, however, be the same if Block 1 is taken as the output circuit and terminals
k, h are, for example, the source and drain of a device which has been equipped with a
specified input network.)
What is the condition that the circuit will be oscillating if terminals g and h, and j and k,
are connected together? Assuming that j, k are already connected, the total impedance
between terminals g, h is Z( f ) = ZG( f ) + Zin( f ). Clearly, when g, h are shorted together,
the circuit will be oscillating if the total impedance Z( f ) obeys the condition already stated
above in connection with Figure 4.9(a).
The following deductions can now be made:

1. The generator circuit, Block 1, has been assumed to be passive, so it can only have
positive resistance. Hence if Block 2 containing the active device only exhibits positive
resistance, the plot of total impedance cannot enter the left-hand half of the impedance
plane and certainly cannot cross the negative axis. Oscillation is therefore impossible.
2. If Block 2 has Rin < 0 at some frequencies, this does not necessarily imply that oscillation
is present. Even if R = RG + Rin remains < 0 at some frequencies, it may be that the total
circuit reactance prevents the Z plot from crossing the negative real axis in the manner
required for oscillation.
3. If Block 2 has negative resistance at some frequency f, then there will always be some
choice of Block 1 that can make the circuit oscillate at that frequency. At a single
frequency, a passive circuit can be designed to have any value of X (positive or negative)
226 Microwave Devices, Circuits and Subsystems for Communications Engineering

and any value of R > O. So RG, XG could be chosen to make RG + Rin = 0, XG + Xin = 0 at
that frequency, at which the circuit would then just be starting to oscillate.
4. If a pure resistance Rd is placed between terminals g and h, then seen at terminal g
there is effectively a new value of Zin with Rd added to the original value. The effect is
to move the impedance plane plot of Zin bodily to the right by an amount Rd. If the
original Rin becomes negative at some frequencies, but the negative values stay finite at
all frequencies (including as f → ±∞), then a sufficiently large value of Rd can make
oscillation impossible for all possible choices of Block 1 (provided, of course, this
remains passive.)
5. If a shunt conductance Gd were connected across terminals h and k, its value would
be added to Gin and this would move the plot of Yin = 1/Zin bodily to the right by
the amount Gd. If any negative values of Gin remain finite at all frequencies including
±∞, a sufficiently low shunt resistance can prevent the admittance plot from entering
the left half admittance plane, thus making oscillation impossible for any choice of
Block 1.

Case (5) is the one which most often arises with typical transistors.

Self-assessment Problem
4.4 (a) Show that replacing Z by −Z in the fundamental Γ – Z relation is equivalent
to replacing Γ by 1/Γ. You can now see that ZG + Zin = 0 is equivalent to
ΓG · Γin = 1.
(b) Prove that an oscillation condition at the input of a two-port network is equiva-
lent to an oscillation condition at the output. This means that if ΓG · Γin = 1,
then ΓL · Γout = 1, and conversely. (It is assumed that S12 and S21 are nonzero.)
This can be proved in a few lines of algebra from Equation (4.12) and the
analogous expression for Γout. The mathematical result is saying what is phys-
ically obvious, that an oscillating system is oscillating everywhere, except
possibly where there is no coupling between two parts of the system.

Additional comments:

1. To really understand the principles of oscillation it is necessary to generalise the idea


of frequency to include the idea of complex frequency. This is likely to seem an
unfamiliar idea unless you are well used to network theory, but a rough intuitive idea of
it will be enough for the time being. Consider the function of time e st. If we make s a
complex number s = σ + jω, then est = eσ t · e jω t. The imaginary s axis (‘real frequency’)
corresponds to the usual idea of frequency as the purely oscillatory function e jω t.
Values of s with σ > 0, i.e. with s in the right-hand side of the s-plane, correspond
to exponentially growing oscillations, and those with σ < 0 to exponentially decaying
ones. All quantities such as impedances, reflection coefficients, etc., that we normally
think of as functions of ‘ordinary’ frequency can in fact be defined over the entire
s-plane.
Amplifier Design 227

The oscillation condition is actually that the total impedance in the circuit is zero, and
equivalently ΓG · Γin = 1, at a value of s in the right-hand side of the s-plane. A zero total
impedance means that a current can exist with no driving voltage, and an exponentially
growing current described by that value of s will occur. (Until of course the system starts
to become non-linear!)
2. You may have come across Nyquist’s criterion for oscillation in a control system with
negative feedback. The criteria used here in terms of impedances or Γs are similar, so
that for example the system will be oscillating if the plot of ΓG · Γin for a real frequency
increasing from −∞ to ∞ encloses the real point +1 in a clockwise sense.

Summarising, the main point is that the possibility of oscillation in simple single-device
circuits always requires the existence of negative resistance, in some frequency ranges, at
either input and output terminals. Avoiding negative resistance will always make oscillation
impossible. When negative resistance does exist, this does not necessarily mean the circuit is
oscillating, but it is a safer strategy not to have practical designs working in this condition.

4.3.2 Production of Negative Resistance


Modern RF design procedures using CAD try to make very accurate predictions, including
many small effects such as ‘stray’ components, line discontinuities, etc. Modelling of devices
often uses elaborate equivalent circuits that aim to reproduce the measured parameters very
accurately. Relying on the computer, and trying to work to high accuracy, we can sometimes
forget how useful it is to find a simple physical understanding of an effect, and how a
drastically simplified model can often reveal the essence of a problem. An example occurs
in the production of negative resistance by transistors. We might know from the measured
s-parameters that a transistor is capable of showing negative resistance, but this would not
reveal the mechanism by which it occurs.
The most basic mechanism producing negative resistance is internal capacitive feedback,
and for MESFETs and other FETs this can be explained very simply. Figure 4.10 shows a
ruthlessly simplified model of a MESFET. Assuming that a source impedance ZG is con-
nected across terminals G and S, the problem is to find the output impedance seen between
drain and source. The method is to apply a voltage VDS and calculate the drain current ID
that results. Even a simple equivalent circuit would normally include a capacitance CDS
from the drain terminal to ground, but this is initially omitted because, at the first level of

Figure 4.10 Highly simplified MESFET model


228 Microwave Devices, Circuits and Subsystems for Communications Engineering

approximation, it would only add susceptance at the output and hence would not affect the
question of the sign of the output resistance.
The applied voltage is coupled via the feedback capacitance to the parallel combination
of CGS and ZG, and develops a voltage across them. This controls the flow of drain current.
Now suppose that the source impedance is a pure inductor L. If the applied frequency
is below the resonant frequency of the L/CGS combination, this combination will appear
inductive. Also if the frequency is kept low enough, we can be sure that the reactance of the
feedback capacitance CDG is greater than the inductive reactance of the L/CGS combination.
The impedance of the whole feedback path (CDS in series with L/CGS) is now capacitive, and
the current in CDG is 90° leading the applied voltage VDS. However, the inductive reactance
of the L/CGS combination will develop a voltage across it that is 90° leading the current
fed into it. Hence VGS has a total lead of 180° with respect to VDS, i.e is in antiphase with it.
The drain current GmVGS is thus in antiphase with VDS and we have an apparent negative
resistance at the output.

Self-assessment Problem
4.5 (a) In Figure 4.10, let Za be the impedance of the parallel combination of the
generator and CGS, Zb the impedance of the feedback capacitor CDG, and YG
the generator admittance. Show that, in the case where gm Za Ⰷ 1, the output
admittance Yout is given by:

gm ⋅ Za
Yout =
Za + Zb
Multiply throughout by Ya = 1/Za to show that this can be re-written as:

 1   1 
Yout = gm   = gm  
 1 + Ya Zb   1 + ( jω CGS + YG )/ jω CDG 

(b) Remembering that gm is real and > 0, show that Yout is a negative resistance if
YG is a sufficiently large inductive susceptance.
(c) A simple improvement of the transistor model is to include a series resistor
RGG′ between the gate capacitor at G′ and the external gate terminal G. The
same expression for Yout can be used if RGG′ is included as part of the generator
impedance. Show that negative resistance can now only occur below a certain
frequency.

4.3.3 Conditional and Unconditional Stability


A two-port network or device is said to be unconditionally stable, at a given frequency f0,
if the real part of its input impedance remains positive for any passive load connected to
its output port. In other words, the device cannot produce a negative input resistance at
frequency f0 when any load with positive (or zero) resistance is connected at the output.
Expressed in terms of Γs, in the notation of Section 4.2.6, the two-port is unconditionally
Amplifier Design 229

stable if | Γin | ≤ 1 whenever any load with | ΓL | ≤ 1 is connected at its output. If the two-port
is not unconditionally stable, it is called conditionally stable or potentially unstable.
The condition could have been stated equivalently that no passive source could give
rise to a negative output resistance. A device that is potentially unstable can definitely be
made to oscillate at the real frequency f0 by selecting any load ΓL that makes | Γin | > 1. A
source can then be chosen so that ΓGΓin = 1 at f0, i.e. the device is just starting to oscillate at
that frequency. Now ΓGΓin = 1 with | Γin | > 1 implies | ΓG | < 1. Furthermore it was already
seen in Self-assessment Problem 4.4 that an oscillation condition at the input is equivalent to
one at the input, i.e. the condition ΓGΓin = 1 implies also that ΓLΓout = 1. Hence we know that
the chosen ΓG is making | Γout | > 1. Therefore a device that is potentially unstable using the
original load-to-input definition could not be unconditionally stable by the source-to-output
definition, and vice versa. The two possible definitions must be exactly equivalent.
If a device is unconditionally stable in a band of frequencies containing f0, it cannot
be made to oscillate at frequencies in that region. When the stable band contains the target
operating band for our amplifier, this is the easiest design regime but it is still essential
to think about what is happening at frequencies far out of the operating band; most single
transistors are potentially unstable at some frequencies.

4.3.4 Stability Circles


The stability circles are an almost indispensable graphical aid in amplifier design. The out-
put stability circle of a two-port, at a given frequency, is the locus in the ΓL (load reflection
coefficient) plane which gives rise to Γin (input reflection coefficient) values lying on the unit
circle | Γin | = 1. (i.e. purely reactive input impedances.) We know that this locus is indeed a
circle because of the general property of bilinear mappings that they map circles into circles.
The stability circle is the image of the Γin unit circle under the bilinear relation, Equation
(4.12), which expresses Γin as a function of ΓL.
The stability circle therefore divides the ΓL plane into the stable region, giving rise
to positive resistance on the input side, and the potentially unstable region giving rise to
negative resistance. If the two-port is unconditionally stable, the potentially unstable region
must have no points lying within the unit circle. This means that either (a) the stability circle
lies entirely outside the unit circle, without enclosing it, and the potentially unstable region
is its interior; or (b) the stability circle encloses the unit circle, and the potentially unstable
region is its exterior. These two cases are illustrated in Figure 4.11.
Another circle that may be considered is the locus of points produced in the Γin plane by
points on the unit circle in the ΓL plane. If the two-port is unconditionally stable, this circle
must lie wholly within the unit Γin plane, and be mapped interior to interior. This is also
shown in Figure 4.11, where different shadings have been used to indicate the regions which
map into each other.
Obviously the input stability circles in the ΓG plane are defined in an exactly analogous
way. Stability circles can be plotted on demand by the various CAD packages. Usually it
will be indicated whether the interior or exterior of the stability circle is the potentially
unstable region. However, if needed, a simple check can be made as follows. The point
ΓG = 0 by definition makes Γout equal to S22. Thus, the origin of the ΓG plane must lie in
the stable region if | S22 | < 1 and in the potentially unstable region if | S22 | > 1. Of course, the
known value of | S11 | can be used in the same way to identify the two regions in the ΓL plane.
230 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 4.11 Typical appearance of stability circles in the unconditionally stable case

4.3.5 Numerical Tests for Stability


It is useful to have a quick numerical test for whether a transistor or other two-port is
unconditionally stable at a given frequency, without having to plot out the stability circles.
This is provided by the stability factor which is conventionally written ‘k’ or less commonly
‘K’, and defined as:

1 − | S11 | 2 − | S22 | 2 + | ∆ | 2
k= (4.20)
2 | S12 || S21 |

The basic test is that k > 1 is a necessary, and usually sufficient, condition for unconditional
stability. The definitions of k and a group of related parameters are probably not worth
committing to memory, but one should be aware of them because they appear in several
important expressions.
The mathematics shows that k > 1 is not by itself an absolute test of unconditional
stability for arbitrary two-ports. In the most general case, it is a necessary but not sufficient
condition, and one or two extra parameters must be checked. Table 4.2 shows three alternative
sets of necessary conditions that can be applied. For example, to use test A, it is necessary
and sufficient for unconditional stability to have k > 1 and | B1 | > 0. B1 and C1 are standard
notations for two further parameters that are important in amplifier theory, and defined in
the table for completeness. (Two further parameters B2, C2 are defined like B1 and C1 but
with ports 1 and 2 interchanged.)
Amplifier Design 231

Table 4.2 Tests for unconditional stability

Definitions:
1 − | S11 | 2 − | S22 | 2 + | ∆ | 2
stability factor k =
2 | S21 | | S12 |
B1 = 1 + | S11 |2 − | S22 |2 − | ∆ |2
C1 = S11 − ∆S*22
∆ = S11S22 − S12S21

Note: ∆, also written D in some texts, is the determinant of the S parameter matrix.

For single transistors, however, the conditions | ∆ | < 1 and B1 > 0 are virtually always
obeyed at all frequencies and it is only necessary to look at the value of k to check stability.
Some transistors will be found to violate the auxiliary conditions (involving | S21 | | S12 |) in test
A at some frequencies, but normally only where k is in any case less than 1.
A major factor in determining k is the denominator. For typical transistors S21 usually
falls with frequency and S12 rises, so that their product is greatest in a mid-range of
frequencies. Thus transistors are usually unconditionally stable at very low and very high
frequencies.
When our two-port in question is cascaded at its output with another two-port network,
such as a matching or stabilising network (of which more presently), we shall use the
notation ΓL for the reflection coefficient seen by the original device and Γ L′ for the value
presented to the second network by the final load. Thus ΓL is a function of Γ L′ defined by
the s-parameters of the intermediate network. The commonest design problem is of course
to choose the second network to obtain the right ΓL when Γ L′ = 0.
Let us define the accessible region in the ΓL plane as all the values that ΓL could take if
Γ L′ is allowed to range over its entire unit circle. A very important property of any strictly
lossless network, provided it has S12 ≠ 0, is that the accessible region at its input is also the
unit circle. In other words, for a truly lossless network, we can obtain any passive (R ≥ 0)
impedance looking into its input with a suitable choice of a passive load impedance. This
makes it obvious that the unconditional stability (or otherwise) of device is unchanged if it
is cascaded with lossless networks at its input and/or its output. In fact, it can be shown that
the stability factor k is invariant under such cascading.
Any one of the three following sets of conditions are necessary and sufficient for uncon-
ditional stability.

A k>1 and | S12 | | S21 | < 1 − | S22 | 2 and | S12 | | S21 | < 1 − | S11 | 2
(Note that the last two conditions also imply | S11 | < 1 and | S22 | < 1.)
B k>1 and B1 > 0 and | S11 | < 1, | S22 | < 1
C k>1 and | ∆ | < 1 and | S11 | < 1, | S22 | < 1

4.3.6 Gain Circles and Further Gain Definitions


Now that stability has been defined, it is convenient to return to discussing the appearance
of gain circles, but we first need to look at the concept of the simultaneous conjugate match
(SCM).
232 Microwave Devices, Circuits and Subsystems for Communications Engineering

Under the SCM condition, the two-port is conjugately matched to both its generator and
its load, so that:

ΓG = Γ *in and ΓL = Γ *out (4.21)

A two-port can be simultaneously conjugately matched if it is unconditionally stable at the


frequency in question. If conditionally stable, oscillation will occur if an attempt to set up
SCM is made.
For a two-port where the feedback S12 is not negligible, the input and output matching
processes are not separable and two simultaneous equations must be solved:

S11 − ∆ΓL
ΓG* = (4.22)
1 − S22 ΓL

S22 − ∆ΓG
Γ*L = (4.23)
1 − S11ΓG

After eliminating one variable, a quadratic equation results which can be solved explicitly.
Following considerable re-arranging of terms, the SCM solution can be reduced to:

B1 + − B12 − 4 | C12 |
ΓG = (4.24)
2C1

B2 + − B22 − 4 | C 22 |
ΓL = (4.25)
2C2

An unconditionally stable device has B12 > 4 | C1 |2 and B22 > 4 | C2 |2. Using the minus signs
gives a solution with | ΓG | < 1 and | ΓL | < 1 which is of course the practically useful one.
Using the plus signs a second solution with | ΓG | > 1 and | ΓL | > 1 is also obtained, and
corresponds to using an active (i.e. negative resistance) generator and load.
Under SCM conditions, GT has been optimised with respect to both ΓG and ΓL (recall
Section 1.1.5), and then becomes equal to both GA and Gp. This peak gain value is called
the maximum available gain of the two-port at the given frequency, and written Gma. By
substituting the SCM values of ΓG, ΓL into the expression for transducer gain, after con-
siderable re-arranging it can be shown that Gma is given by:

| S21 |
Gma = ⋅ (k − k 2 − 1 ) (4.26)
| S12 |

Gma may be defined as the highest gain that the two-port device can possibly achieve at the
frequency in question, subject to the proviso that no feedback path exists between the input
and output networks. The gain could, in principle, be increased by introducing positive
feedback. This was common in early radio receivers but is never done nowadays. Gma can
only be defined where the two-port is unconditionally stable at the given frequency.
It is now possible to describe the general appearance of gain circles. Taking first the
unconditionally stable case, we consider the circles of constant GA in the ΓG plane. The
expression derived in Self-assessment Problem 3 (b) shows that GA = 0 (−∞ in dB) on
Amplifier Design 233

the unit circle. We therefore see a set of nested (obviously non-intersecting) circles such
that GA is zero on the unit circle. As the constant value of GA is increased from zero, the
circle shrinks and converges on the single point at which the maximum value Gma occurs,
and at which ΓG is given by Equation (4.19). The centres of the circles all lie on the diameter
of the circle which passes through this point. The general appearance of Gp circles in the
ΓL plane, also converging on the SCM point, is similar.
If the two-port is potentially unstable, the gain circles have a different appearance,
whose general form is shown in Figure 4.12. This diagram shows the mappings: ΓL ↔ Γin in

Figure 4.12 Typical appearance of gain and stability circles for potentially unstable two-port
234 Microwave Devices, Circuits and Subsystems for Communications Engineering

the top half of the diagram, and ΓG ↔ Γout in the lower half. In fact, it is slightly neater to
use the Γ*in and Γ*out planes, which are simply the Γin and Γout planes reflected about the
horizontal axis.
The ΓL plane unit circle is mapped into a certain circle (its ‘image’) in the Γ*in plane, and
the unit circle in this plane maps to the stability circle in the ΓL plane. Because the device is
potentially unstable, we know that the stability circle and the other image circle intersect
the unit circles, and the intersection points A, B map into the points C, D. A similar set of
relationships exist in the lower half of the diagram.
Interestingly the intersection points A, B turn out to be the same as points E, F, and C, D
the same as G, H. These points are ‘pathological’ points which obey both the oscillation
condition and the SCM. They are the SCM solutions given by Equations (4.19, 4.20) in the
potentially unstable case where B 12 < 4 | C1 | 2 and B22 < 4 | C2 | 2. At these points we have, for
example, ΓG = Γ*in, but this also implies ΓG = 1/Γin or ΓG Γin = 1 (the oscillation condition)
since | ΓG | = 1.
It is now easy to deduce the form the gain circles must take. In the lower left part of
Figure 4.12, the section of the unit circle outside the potentially unstable region is easily
shown from the GT expressions to have GA ≡ 0 (−∞ dB), so is itself one of the gain circles.
The non-zero GA circles must not intersect either this, the stability circle, or each other, and
it is clear that the only way they can be fitted into the space between the unit circle and
the stability circle is if they all pass through the pathological points as shown. We see the
gain circles moving towards the stability circle as GA increases. GA becomes infinite on the
stability circle and in the potentially unstable region. At the pathological points, the circle
of infinite gain intersects the circle of zero gain. Gain becomes undefined at these points
but can take any value in their neighbourhood. The design points should obviously be kept
well away from them.
Similar comments apply to the Gp circles in the top right of Figure 4.12. Although Gp
becomes infinite on the stability circle, it actually has finite but negative values (not negative
dBs, but negative numbers) within it. This means simply that a load in this region produces
negative resistance at the input side, and there is consequently a net power flow back into
the generator as well as out into the load. It is not recommended to operate in this condition
but it is possible in principle.
This section concludes with two final definitions of special power gain quantities. It
is normally possible to resistively load a potentially unstable two-port device, at its input
and/or its output, to make the whole network become unconditionally stable. (This would
only be impossible where the potentially unstable region enclosed the entire unit circle,
which would not happen for normal transistors.) Imagine the original two-port, cascaded
with any loading networks, as a single new two-port, and suppose the loading is adjusted
to the point where this new two-port just becomes unconditionally stable. The maximum
stable gain of the original two-port, written Gms, is defined as being the value of Gma for the
entire network, and is a useful figure of merit. Gms can be defined as the largest gain that
the potentially unstable device (again disallowing feedback) can give under the requirement
that the generator and load must be simultaneously matched at the outermost reference
planes and this can of course only be achieved by the introduction of loss between the
original two-port and at least one of the outer reference planes.
The resistive loading to the point where the network is just unconditionally stable makes
k = 1 and the above expression for Gma then shows that Gms would be given by the ratio
Amplifier Design 235

Figure 4.13 (a) Transistor with general lossy output loading; (b) a specific example

Figure 4.14 Accessible region for ΓL with series resistive loading

S21/S21 for the entire loaded network. In fact, this remains the same as S21/S21 evaluated for
the original two-port itself.
It is important to realise that not all loading topologies can produce the stability in any
given case. Consider Figure 4.13(a) where the two-port (in this case shown as a transistor)
is cascaded with a lossy network at its output. ΓL is the reflection coefficient seen by the
transistor output and Γ′L is that presented by the external load at the outermost reference
plane. If the lossy network is just a series resistance (R) as in Figure 4.13(b), connecting any
passive load can only give a ΓL corresponding to a larger value of resistance. The region
accessible to ΓG is therefore the interior of the circle of resistance R in the Smith chart,
Figure 4.14. As G is increased from 0, this circle will shrink down towards the open circuit
point ΓL = +1. If the device’s region of potential instability were B, no amount of series
resistance would prevent the accessible region from intersecting the potentially unstable
236 Microwave Devices, Circuits and Subsystems for Communications Engineering

region. However, if the potentially unstable region were A, a sufficiently large R would
make the regions non-intersecting, thus producing unconditional stability.
One further gain quantity is theoretically very important. This is the unilateralised gain,
usually written U. At a single frequency, a two-port can always be unilateralised, i.e. pro-
vided with a lossless feedback network which will cancel its own internal feedback,
making S12 of the whole network equal to zero. (A practical method of doing this called
‘neutralisation’ was used quite frequently in the past, but is not now used in low power
transistor amplifiers.) U is defined as the value of Gma of the whole unilateralised network,
and can be shown to be given by:

1/ 2 | S21 / S21 − 1 | 2
U= (4.27)
k | S21 / S12 | − Re(S21 / S12 )

(Caution: U is not to be confused with GTU, the last entry in Table 4.1, which is simply an
estimate of GT obtained by neglecting S12, or with GTUmax, the value of GTU maximised by
setting ΓG = S*11, ΓL = S*
22. U would equal GTUmax for a device that actually had S12 = 0. Note
also that some books use symbol U for a related quantity called ‘unilateral figure of merit’,
which gives an indication of how good the approximation of neglecting S12 is.)
U is not a very useful parameter in circuit design procedures, but is important in
characterising devices. U turns out to be invariant if the two-port device is embedded in
any lossless network whatsoever, provided two ports are still presented to the outside world.
This embedding also includes change of the common terminal, so that a transistor in the
usual common emitter connection has the same U if operated in common base configura-
tion, etc.
The condition U > 1 is the condition that a device is ‘active’, i.e. capable in principle of
giving power gain if surrounded by sufficiently ingenious networks. A device with U < 1
is incapable of giving any power gain no matter what circuitry surrounds it. For any device
U must fall to zero at some sufficiently high frequency fmax, which is called the maximum
frequency of oscillation of the device. For transistors U decreases continuously with frequency,
though other more exotic amplifying devices might also have U > 1 only above a minimum
frequency, or even in more than one separate frequency band. Notice that if we had a device
unilateralised to make S12 = 0, we could also equip it with matching networks so that we
could realise a device with S11 = S22 = S12 = 0, | S21 |2 = U. If U < 1 it appears physically
obvious that there is nothing we could do to make this network give any gain, but if U > 1
it is obvious that an oscillation condition could be created just by connecting a cable of
suitable length, and impedance equal to the reference impedance Z0, from its input to its
output. Hence the name given to fmax, which is clearly the highest frequency at which the
device could possibly be made to oscillate.
In conclusion, it would be useful to inspect Figure 4.15 showing the typical frequency
variations of gains for a typical MESFET. Notice that the device becomes unconditionally
stable above about 7 GHz, and the Gma curve definable above this frequency is continuous
with the Gms curve below it. fmax is 80 GHz, and it is interesting to note from the U curve that
significant gain could still be obtained at about 45 GHz where Gma has fallen to unity (0 dB).
This would be rather hard to do in practice, as it would require some kind of feedback
network, and it would be easier to find a higher frequency transistor.
Amplifier Design 237

Figure 4.15 Dependence of gain on frequency for typical MESFET

Self-assessment Problem
4.6 In connection with resistive loading of a two-port to make it unconditionally
stable, we considered the accessible regions for ΓL under series resistive loading.
What would the accessible region be if the loading network were:
(a) a shunt resistor of conductance G?
(b) a matched 10 dB attenuator, i.e. with S11 = S22 = 0, | S21 | = | S12 | = 1/√10?
Could either or both of these loading types produce unconditional stability if the
potentially unstable region were B in Figure 4.14? How could you realise the
10 dB attenuator as a T-network of three ideal resistors?

4.3.7 Design Strategies


We are now in a position to summarise some actual design procedures for designing an
amplifier to give useful transducer gain at a single target frequency f0, leaving aside for
the moment the issues of optimising its noise performance or obtaining substantial band-
width. The design problem is as follows: given values Γ′G, Γ′L (most commonly zero) for the
generator and load reflections that our amplifier will see at its interfaces with the outside
world, we must design matching networks between these interfaces and the active device’s
ports so that these are transformed into the correct ΓG, ΓL as seen by the device.
One useful point to note is that for the circuit to be oscillating, both ΓG and ΓL must be
in their potentially unstable regions, and the device must be showing negative resistance at
both ports. It was shown earlier that an oscillation condition at the input is equivalent to
an oscillation condition at the output. Thus, if (e.g.) ΓL were in its stable region, the input
resistance would be > 0 ( | ΓL | < 1), no oscillation would then be possible at the input, and
238 Microwave Devices, Circuits and Subsystems for Communications Engineering

hence none would be possible at the output – even if ΓG were in its potentially unstable
region producing a negative resistance on the output side. Thus, it is only necessary to
ensure one of ΓG and ΓL is in its stable region to prevent oscillation. It can thus be seen that
operation without oscillation, yet with negative resistance present on one side, is possible in
principle, though not very desirable.
The following procedures can be applied:

1. If the device is unconditionally stable at f0:


(i) Select ΓG, ΓL for a simultaneous conjugate match at f0; this not only maximises the
transducer gain but also ensures that the generator and load will also see a match
when looking into our designed amplifier. Nominally lossless matching networks
can be used and GT can be made equal to Gma.
(ii) Ensure that at all frequencies where the device is potentially unstable, normally in
a band somewhere below f0, at least one of ΓG, ΓL (preferably both) is outside the
potentially unstable region, to prevent oscillation.
(iii) If Γ′G and Γ′L are not sufficiently predictable in the potentially unstable bands to be
sure of meeting condition (ii), consider introducing resistive loading networks. These
can be designed so that resistive damping only appears in the potentially unstable
band. For example, a shunt resistor might be connected from gate/base to ground,
and placed in series with a parallel tuned circuit or quarter wave resonant line which
will remove the damping in the neighbourhood of f0. Alternatively, it might be
placed in series with an inductor which would effectively remove its influence at all
sufficiently high frequencies.
2. If the device is conditionally stable at f0, then:
(i) a possible strategy is to load it resistively, using an appropriate topology as dis-
cussed in the previous section, to the point where it becomes unconditionally stable,
then proceed as in (1). This has the advantage that the outside world can see an
impedance match looking into both of the outermost ports, i.e. at the reference
planes which enclose the resistive loading and the matching circuits we have added.
It also has the advantage that oscillation cannot be produced if the Γ′G and Γ′L at
f0 should happen to be a long way from their nominal values. The maximum trans-
ducer gain is limited to Gms and there would be some penalty in the saturated power
handling with output resistive loading, or noise figure with input loading.
(ii) If alternatively lossless matching networks are retained, an impedance match will
not be possible at both of the outermost ports, but can still be obtained at either the
input or the output.
(iii) Suppose that the choice is made to have a match at the input. The operating power
gain (Gp) circles (Figure 4.12) can be plotted to find a value of ΓL which is outside
the potentially unstable region and not too near the unit circle. The resulting Γin
can then be calculated from Equation (4.12), and the input matching circuit can be
designed to make ΓG conjugate matched to it. The indicated Gp value will then of
course be the GT that is actually achieved.
(iii) In principle, the Gp selected can be arbitrarily large, by selecting ΓL close enough to
the stability circle. In practice an upper limit on the selected Gp can be imposed by
the requirement that the ΓG required to perform the input match also remains outside
its potentially unstable region. It is also possible to set a limit by considering the
Amplifier Design 239

likely tolerance on the ΓL that will actually be presented (because of the tolerance
on Γ′L relative to its nominal value, and manufacturing tolerances in the matching
circuit that is actually made) to ensure that the actual ΓL in practice can never move
into its potentially unstable region. Finally, a limit on Gp might be imposed when the
design is required to have substantial bandwidth – too large a value will make the
required bandwidth impossible to obtain.
(iv) The same procedure can be used making the alternative choice that a match is to
exist at the output, and then using the GA circles and stability circle in the ΓG plane.

In all cases the circuit will present mismatches outside its operating band at both of the
interfaces with the outside world. In case (2),(i) the circuit must also present a mismatch
to the generator within the operating band, or likewise to the load in case (2),(ii). These
mismatches may cause unpredictable gain or gain variation with frequency, or possibly
oscillation, when the circuit is interfaced to external circuits that are also imperfectly matched.
The balanced amplifier technique (discussed later) is a useful way of avoiding these problems.

4.4 Broadband Amplifier Design


The broadband design problem arises when an amplifier or other radio frequency component
is required to work over a band of frequencies that is a substantial fraction of the frequency
at the centre of the band. (This is commonly expressed as a ‘percentage bandwidth’.) A
design made for a single frequency will usually give at least a few percentage of useful
bandwidth, over which there is negligible gain variation, without making any further effort.
In a broadband design the transducer gain must have a specified variation over a substantial
percentage bandwidth. By far the most common case is that the variation should be minimised,
i.e. the gain should ideally be ‘flat’ over the desired band.
There are four main techniques for achieving broadband operation:

1. negative feedback;
2. distributed amplifiers;
3. compensated matching.
4. special connections of two or more transistors as a unit.

The compensated matching technique simply consists of designing for a controlled degree of
mismatch at the input and/or output which is designed to vary in such a way as to compensate
for the inherent tendency of the active device’s gain to vary (usually fall) with frequency.
Associated with these methods is another, the balanced amplifier. This is not strictly a
broadbanding technique in its own right, but rather one which reduces the interactions of the
amplifier with the outside environment, and hence makes the compensated matching approach
easier to apply.
The distributed amplifier is an important method for achieving extremely high (multi-
octave) bandwidths. The price of its high performance is that many active devices are
used in the one amplifier. This specialised method is outside the scope of this discussion,
which is restricted to methods that could be used to improve the bandwidth achievable with
a single device. There are also arrangements where a small number of transistors may be
connected together as a single unit.
240 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 4.16 Scattering parameters of the ATF-36163 FET

4.4.1 Compensated Matching Example


Taking an example transistor, ATF-36163, its s-parameters are as shown in Figure 4.16. The
S21 of the transistor gradually decreases with frequency while S11 is always greater than S22,
which naturally has a good match near 8 GHz. We can then add reactive matching networks
to reduce the reflections and therefore increase the S21 of the complete circuit. The equations
in Table 4.1 show that in the unilateral case the gains of the input and output networks have
a maximum value (when conjugately matched) of 1/(1 − | S | 2 ). The greatest reflection, S11 in
this case, will produce the greatest increase in gain when matched out.
Requiring the input network to produce a match at 11 GHz and the output network to
match at 12 GHz the complete circuit response becomes that shown in Figure 4.17. Clearly

Figure 4.17 Scattering parameters of the ATF-36163 FET with matching circuit
Amplifier Design 241

the removal of the reflections in the 10 GHz to 12 GHz region has increased S21 in that
region, producing a relatively flat profile up to 12 GHz, but then drops off quite sharply.
Such an amplifier is well matched at the frequencies where we have compensated for the
natural roll-off of the transistor, but is poorly matched at the lower frequencies.

4.4.2 Fano’s Limits


Design of amplifiers and other components for broadband operation depends on being able
to perform impedance matching over wide ranges of frequency, and this operation proves to
be subject to fundamental limits. Consider the situation shown in Figure 4.18(a), where a
lossless, passive matching network is to be used to match a given load to a purely resistive
generator. Γ will denote the input reflection coefficient using the source resistance R0
(usually 50 Ω) as the reference impedance.

Figure 4.18 Topologies for defining Fano’s limits


242 Microwave Devices, Circuits and Subsystems for Communications Engineering

If the load is purely resistive, Γ can be made as small as desired over an arbitrarily large
frequency range, even if the load resistance is not equal to R0. (In principle, an ideal trans-
former would do this; other methods such as a transmission line with a gradually tapered
or multi-stepped Z0 are more practical at UHF and above.) If the load includes reactive
components, it can be shown, as one would expect intuitively, that Γ cannot be made as
small as we like over unlimited bandwidths.
For certain simple loads, shown in Figure 4.18 (b) to (e), the theoretical limits on match-
ing have been found, and these are known as Fano’s limits:

冮 ln  | Γ1 | dω ≤ πτ
0
for load topologies (b), (c) (4.28)

冮 ω1 ln  | Γ1 | dω ≤ πτ
0
2
for load topologies (d), (e) (4.29)

τ is the time constant of the load, given by RC or L/R as appropriate. To understand


the meaning of one of these expressions, notice that the term ln(1/| Γ |) in the integrand is a
measure of the ‘goodness’ of the matching scheme at the frequency in question. Smaller
values of | Γ | make the integrand larger, zero values make it infinite, and a complete mis-
match with | Γ | = 1 makes it zero. The integrals are saying that the total ‘goodness’ integrated
over frequency cannot exceed a certain limit set by the load. Thus if we attempt to improve
the match in certain frequency ranges, we must expect it to worsen at others. Since the
normal design problem is to produce a system which functions just over some specified
frequency band, and the performance out of band is immaterial, it is clearly a good strategy
to make (ideally) | Γ | = 1 outside the operating band, so we can have the maximum ‘goodness’
within it.
It is also immediately obvious from the expressions that no matching scheme can make
| Γ | = 0 over any finite range of frequency, as this would automatically make the integral
infinite, but it is possible to have | Γ | = 0 at a finite number of isolated points on the
frequency axis, and in this case the integral will converge to a finite value even though the
integrand is tending logarithmically to infinity at these frequencies.
The first two topologies (b) and (c) clearly have the property that they become more
difficult to match over wide bandwidths as the time constant of the load network increases;
in the limit where the capacitor or inductor tend to zero, they just become pure resistors
again. On the other hand, circuits (d), (e) become easier for broadband matching as the time
constant increases; for example, in (d), as C → ∞ it looks like a short circuit and reduces the
load to a resistor again. This is an easy way to remember whether the time constant appears
on the top or bottom of the right-hand side of the equation: for example, for circuit (b), the
integrated goodness must be smaller for a larger time constant, which therefore appears on
the bottom of the right-hand side.
It is also easy to check whether the integral should include the 1/ω 2 factor by looking at
the dimensions of the expressions. For example, in Equation (4.28), the logarithm term is a
pure number, and the dimensions of the integral are those of dω, namely, a frequency or
equivalently a reciprocal of time, and this is also the form of the right-hand side.
Amplifier Design 243

Self-assessment Problem
4.7 Show that the dimensions are also correct for the other form of the integral con-
taining the 1/ω 2 term.

No attempt will be made here to prove the limits formally. However, the issue can be simplified
a little by pointing out that, although there appear to be four cases of the limit, there is really
only one. If one of the limits is taken as given, the others could all be deduced from it by
duality arguments. This is explained in a Self-assessment Problem at the end of this section.
The results can also be made more plausible by pointing out each integral is in fact exactly
equal to the limit value if the matching network is omitted entirely, provided the resistor
in the load is equal to the source resistance. (This of course makes the match perfect at
either zero or infinite frequency.) This result is easy to prove by a straightforward integration
of the expression for Γ for the no-matching case. Hence we can have no more ‘integrated
goodness’ than is possible for no matching device at all; the best the matching can do is to
re-distribute the goodness to frequencies where it is more useful to us.
It has already been mentioned that compensated matching is one of the most important
techniques for broadband amplifier design. The tendency of transistor gain to fall with
frequency is compensated by making the input and/or output mismatch worse at lower
frequencies. When this technique is used, the need to intentionally create controlled mis-
match eases the demands of the Fano limits.
The limits can be applied where an approximate model for the device exists, and they really
need the approximation that the device is unilateral. Fortunately this usually holds reasonably
well for high frequency problems. Figure 4.19 shows the very simplified model of a GaAs
MESFET that could be used for a rough assessment of the gain bandwidth limitations.
In this simplified model we see that the input network is of type (d) in Figure 4.18 while
the output network is of type (b).

4.4.3 Negative Feedback


Negative feedback is a familiar technique in low frequency amplifiers where it offers the
benefits of:
1. gain and phase responses that have less frequency variation;
2. reduced non-linear distortion;
3. performance that is less sensitive to variations between individual samples of the active
devices, because it is at least partially controlled by external passive components.

Figure 4.19 Simplified MESFET model for assessing the Fano limits
244 Microwave Devices, Circuits and Subsystems for Communications Engineering

These principles have long been applied in high fidelity audio amplifiers, and they are
developed to their fullest extent in the vast number of precision circuits based on operational
amplifiers.
The same benefits can often be obtained in radio frequency amplifiers, and with the added
advantage that it may be possible to produce a good input and output match to a standard
impedance (usually 50 Ω) over wide ranges of frequency.
However, three basic problems reduce the applicability of negative feedback in RF
amplifiers, especially at the higher microwave frequencies:

1. The amplifier noise figure is usually degraded.


2. The phase of a transistor’s S21, which is 180° at low frequency lags increasingly in phase
as the frequency is raised. Transistors in the common emitter or common source con-
figurations produce phase inversion at low frequency, i.e. the phase angle of S21 is 180°.
This makes it easy to provide negative feedback by a simple shunt or series resistor.
However, as the frequency is raised, the phase departs from 180° and can eventually
reach 90° where the feedback starts to be positive (destabilising) rather than negative.
3. Negative feedback is most effective when it is made to produce a large reduction in the
gain of a device whose intrinsic gain is high. This condition is not met where the
operating frequency is so high that the transistor’s Gma is approaching unity.

This can only be a short introduction to the theory of negative feedback in RF amplifiers.
The conclusion can be summarised that negative feedback is worth considering where some
noise figure degradation can be tolerated, where the required bandwidth is large, and where
the upper limit of the require operating frequency range is either below or comparable to
the turn-over frequency at which S21 moves into the first quadrant.

4.4.4 Balanced Amplifiers


The 3 dB couplers are required to be quadrature couplers. An input at Port 1 produces out-
puts at 2 and 3 only, and the output at 2 is 90° behind that at 3, at the operating frequency.
Of course, for 3 dB coupling these outputs are also equal in magnitude.

Figure 4.20 Basic balanced amplifier configuration


Amplifier Design 245

A lossless matched reciprocal coupler with this property must have the same property for
an input at 4, which must divide equally between Ports 2, 3 only, with the output at 3 being
90° behind that at 2. Thus, in the diagram, the loop Ω on a coupling path represents an
excess phase lag of 90°.

4.4.4.1 Principle of operation


Two nominally identical amplifiers are used then, if the coupler is ideal, Γ = 0 at Port 1
regardless of the ΓIN of the amplifiers. With an input at 1, the wave incident on amplifier 2
is 90° behind that incident on 1. On retracing its path back to Port 1, it receives an additional
excess lag of 90°. Therefore for equal reflections at the amplifiers, reflections at Port 1 are
180° out of phase and cancel (all reflected power appears at Port 4). On the output side, all
the output appears on Port 4 and there is nominally a perfect match.

4.4.4.2 Comments
1. Typically couplers also have the property that couplings from 1 → 3 and 2 → 4 are
of equal phase (arg S31 = arg S42); in this case the coupler at either side could be flipped
about a vertical line. With 3 on the left, 1 on the right, 2 on the right and 4 on the left, the
arrangement would still work.
2. Basic coupled lines, properly matched, have the required isolation and quadrature property.
The coupling ratio, however, is frequency dependent and it is also hard to realise 3 dB
coupling in simple microstrip structures.
3. The Lange coupler is a most common realisation for balanced amplifiers. This is an
improved microstrip coupler giving the required high coupling (3 dB) and quadrature
properties over greater bandwidths – an octave or more. Realisation on a circuit board
can, however, be problematic.
4. A possible alternative coupler is a Wilkinson divider with a + 90° excess path in one arm
(Figure 4.21). This gives a narrower bandwidth, but is simple.

Figure 4.21 A Wilkinson divider with additional phase lag in one arm
246 Microwave Devices, Circuits and Subsystems for Communications Engineering

4.4.4.3 Balanced amplifier advantages


1. The balanced configuration gets rid of reflections, and therefore interactions between
stages. It is then possible to adopt a simple ‘building block’ approach to system design by
cascading such elements together. Nevertheless the active devices in the complete amplifier
can see the mismatches that are needed to realise gain flatness, stability, noise optimisation,
or optimised saturated output.
2. 3 dB more output power at saturation.
3. Intermodulation products are about 9 dB down in the balanced configuration as compared
with an unbalanced equivalent working at same total output power.
4. Greatly improved stability.
5. If one amplifier fails, system may still continue working, with about 6 dB less gain
(provided it does not oscillate).

4.4.4.4 Balanced amplifier disadvantages


More board space and more power are required.

4.5 Low Noise Amplifier Design


4.5.1 Revision of Thermal Noise
This section begins with a brief review of the fundamentals of thermal noise generation. This
should be familiar to readers who have already studied electronics to degree level, but may
still be useful revision. Probably you will have been introduced to thermal noise and shot noise
as fundamental processes producing noise in electronic circuits. We are concerned here with
additive noise, i.e. the random fluctuations added to the signals we are trying to amplify or
otherwise process, and which degrade their information content. Noise can be generated by
several other processes, both within electronic circuits and externally. For example, a signal
collected from an antenna may be accompanied by both thermal and non-thermal noise
generated in the external environment, which includes both the earth and outer space.
It is common to specify the level of noise at any point in a system, regardless of its actual
origin, in terms of an equivalent thermal generator. It is therefore necessary to know how
much thermal power a device at a given temperature can generate.
Any passive device with a pair of terminals (i.e. a passive one-port) has a random noise
voltage across its terminals arising from the thermal agitation of the charge carriers (mainly
electrons) within it. If a load is connected to the device, some noise power can be collected
from it. The fundamental property is that the spectral density of the available thermal power
from the device, i.e. the power which could be collected at a given frequency from the
device by a matched load, is given by:
hf
S( f ) = Watts Hz−1 (4.30)
e hf / kT
−1
where S( f ) denotes the power spectral density, T is the physical temperature of the passive
device, h is Planck’s constant and B is Boltzmann’s constant. The values of the constants are:
h = 6.626 × 10−34 Joule second k = 1.381 × 10−23 Joule Kelvin−1 (both to 3 d.p.)
Amplifier Design 247

Figure 4.22 Equivalent circuits for thermal noise: (a) Norton; (b) Thévenin

Note carefully the units of S( f ), which is a power per unit bandwidth. A random signal has a
continuous spectrum in which the average power collected in the neighbourhood of a chosen
frequency is directly proportional to the bandwidth over which power is accepted, provided
this bandwidth is kept small enough. This is the meaning of the term spectral density.
The full expression for S( f ) may look unfamiliar because it contains the quantum theory
correction. The graph of S( f ) against frequency is virtually flat at low frequencies, then it
falls off very rapidly at frequencies where the quantum energy hf becomes comparable with
the characteristic thermal energy kT. In almost all radio and microwave engineering, hf is
very much less than kT, and the expression for S( f ) can be simplified to a more familiar
form, which is usually an extremely good approximation:
S( f ) = kT Watts Hz−1 (4.31)

The approximate expression will be used here. The exact expression occasionally needs to
be used for more exotic applications, such as remote sensing or radio astronomy at very high
millimetric frequencies, and where very low temperatures are involved.
It is often useful to have Norton and Thévenin equivalent circuits for the thermal noise
fluctuations, as shown in Figure 4.22.
In both cases the real passive device is replaced with a noise-free equivalent which has
the same impedance Z (at the frequency f in question), this being written as Z = R + jX or
as Y = 1/Z = G + jB. In the Norton form (a) the noise is generated in a separate constant
current source, with a random noise current in(t) which is also the short-circuit terminal current.
Likewise in (b), the random noise voltage vn(t) is the open-circuit terminal voltage.
Si( f ) and Sv( f ) will denote the spectral densities (i.e. mean square values per unit band-
width) of the current and voltage generators in these equivalent circuits, and are measured in
Amps2/Hz and Volts2/Hz respectively. They are given by:
Si( f ) = 4kTG, Sv( f ) = 4kTR (4.32)
(Note that, in connection with noise, the term ‘passive’ as applied above to a passive one-
port, must be used in a strict sense to mean a device which has no external power supplies
to it. It must have no inputs of energy other than heat absorbed from its surroundings. An
example would be a resistor. The term passive is used elsewhere in a wider sense to mean
incapable of giving power gain, so that for example a transistor becomes passive in this
sense at frequencies above its maximum frequency of oscillation.)
248 Microwave Devices, Circuits and Subsystems for Communications Engineering

Self-assessment Problem
4.8 Using the approximation ex ≈ 1 + x for small x, derive the approximate expression
S( f ) = kT from the exact one, Equation (4.30).

4.9 If T = 3 K, show that hf = kT at f = 63 GHz. Comment: T = 3 K, the temperature


of the cosmic microwave background radiation, is the lowest temperature we can
ever see in the environment, and well below the temperatures usually encountered.
At this very low temperature, the quantum correction becomes noticeable at a
fairly high microwave frequency. 63 GHz is not too exotic a frequency – 94 GHz
radar systems are now quite common, for example.

4.10 Explain why a factor of 4 appears in Equation (4.33). (This is not specifically a
question about noise, but applies to any generator. The available power from a
generator, whose rms open circuit voltage is V and internal resistance is R, is
given by V 2/4R.)

4.5.2 Noise Temperature and Noise Figure


Two quantities commonly used to characterise the noise performance of two-port devices,
particularly amplifiers, are the noise temperature and the noise figure of the device. These
are simply related and contain equivalent information. The noise temperature of the two-port
device, which will be written Te (for equivalent temperature), is simply a convenient way
of referring the noise generated in the two-port to an equivalent thermal noise generator at
its input. It is defined as follows: let the two-port be connected to a specified generator,
which itself may be producing signals and noise. At the output of the two-port, noise power
is available of which some is generated by the two-port itself.
Now imagine that the two-port is replaced by a noise-free equivalent which is otherwise
identical, and that the generator is replaced by a passive device of the same impedance.
Then if the noise temperature of the real two port Te is defined as the physical temperature,
the generator impedance would have to produce the same noise power at the output as is
actually produced by the internal noise sources of the two-port. This rather cumbersome
verbal definition could just be summed up in the equations:

Sout = Gt kTe (4.33)

where Sout is the spectral density of noise delivered to a specified load, due to the internal
noise sources of the two-port only, GT is the two-port’s transducer gain, and Te is its noise
temperature. We could remove dependence on the load by writing:

SA,out = GAkTe (4.34)

where SA,out is the available noise output spectral density, i.e. what could be extracted by a
matched output load. If we include also the output noise due to the real generator, it would
read:
Amplifier Design 249

SA,out(total) = GAk(Te + TG) (4.35)

where the real generator has been an assigned an equivalent temperature TG such that kTG
is the available noise spectral density from its output port – whether or not that noise is
actually generated thermally.
It is very important to note that Te of a two-port depends on the source impedance from
which its input is fed, but does not depend in any way on the loading at its output.
The noise figure of the two-port will be written F and it quantifies the idea that it degrades
the signal to noise ratio (SNR) of the signal we pass through it. The two-port gain is GA,
linking the input and output signal power. The most familiar definition of noise figure is
usually stated:

SNRin P G k (T + TG ) T
F= = in × A e =1+ e (4.36)
SNRout kTG GA Pin TG

The problem with this definition is that it only makes sense if the absolute noise power
level accompanying the input signal is specified. We could have the same S/N ratio with
different absolute levels of the input noise from the generator and of course at higher
input noise levels, the added noise in the following two-port would have less effect on the
S/N ratio.
To make the definition meaningful, the input noise level must be specified, and unless
otherwise stated it is usually taken as corresponding to a standard room temperature. This is
written T0 and is usually taken as 290 K. In other words, this amounts to assuming TG = T0
in Equation (4.35).
Although F seems to have the advantage that it gives a quick calculation of the degrada-
tion of S/N ratio, this simplicity is often lost because, more often than not in practical
situations, the input noise level does not actually correspond to T0. This is particularly
true of antenna noise temperatures, which can be far above T0 at low frequencies and well
below it in the low noise region of the microwave spectrum between about 1 and 10 GHz.
The noise temperature is in many ways a more elegant definition, but the use of F is very
entrenched. (F is usually given in dB, but remember to convert back to an ordinary number
when the calculations need it!)
A very important expression gives the noise temperature of a cascade of two-ports, in an
obvious notation:

Te2 Te3
Te(total ) = Te1 + + +... (4.37)
GA1 GA1GA2

The interesting feature of this expression is that it can be expressed purely in terms of avail-
able power gains. However, the impedance matching at the output of stage 1 is important,
although it does not affect Te1 or GA1, because Γout,1 affects the second term via its effect on
Te2 (And, if significant, the third term via GA2.) Typically in a well-designed low noise system
the first term should be dominant, the second term small and the third term negligible.
Both high gain and low noise figure contribute to the ‘goodness’ of an amplifier. A
quantity called the noise measure M of a two-port is occasionally useful as a single figure of
merit that includes both factors. M is defined as the noise figure of an infinite cascade of
250 Microwave Devices, Circuits and Subsystems for Communications Engineering

identical copies of the two-port in question. It is useful when all the stages in a receiver that
contribute significantly to F are based on the same active device, or at least have similar
noise figure and gain, but this is often not the case.

Self-assessment Problem
4.11 For three two-ports in cascade, show that the available noise power spectral
density at the final output is kTe1 · GA1 · GA2 · GA3 + kTe2 · GA2 · GA3 + kTe3 · GA3.
How would you deduce the expression for Te,total from this? How would you
write the cascading formula in terms of noise figures?

4.12 Show that the noise measure M of a two-port is given in terms of its noise figure
and gain by M = (GAF − 1)/(GA − 1). (Use 1 + x + x 2 + x 3 + . . . = 1/(1 − x),
provided | x | < 1.)

4.5.3 Two-Port Noise as a Four Parameter System


To describe fully the noise properties of a two-port at a single frequency, four real para-
meters are needed. In the following sections the reason for this will be explained, and will
be used to deduce a general form for the dependence of a device’s noise temperature on
the impedance of the generator that feeds it.
The fundamental point is that the internal noise processes in a two-port cause random
noise waveforms to appear at both its ports. We can in fact extract noise power from both its
input and its output. Furthermore, the noise at the two ports can arise in the same internal
physical process which is coupled out in different ways to both the ports. Hence there is
usually some degree of statistical correlation between the noise waveforms produced at the
two ports.
If we had a random waveform x(t), you will have some idea of characterising it by its
frequency spectrum, and you should understand that this can be described precisely by its
spectral density. However, if we had two (real) random waveforms x1(t), x2(t), what informa-
tion would we need then? Obviously we would need to specify their individual spectral
densities, but this would only be enough when they are completely uncorrelated statistically,
as would happen if they arose from physically separate apparatus. Where some correlation
exists, another quantity called the cross-spectral density must also be specified.
This is a complex quantity, so it consists of two real numbers which together with the
individual spectral densities make up the four real parameters needed to specify the two
noise generators at a single frequency.
To explain the idea of the correlation, imagine that the random waveforms x1(t), x2(t) are
each passed into a filter whose passband is centred on the frequency f in question, and which
has a small bandwidth δ f which is very much less than f. At the output of each filter, we
would see a waveform which, on a short time scale, would look like a sinusoid at frequency
f. On a longer scale, the amplitude and phase of this sinusoid would be seen to be varying
randomly. We could write:
Amplifier Design 251

x1(t)filtered = √2Re(X1(t) · e j2π ft), x2(t)filtered = √2Re(X2(t) · e j2π ft) (4.38)

where X1(t), X2(t) are random complex numbers or phasors (expressed as RMS values)
which vary on a longer time scale of 1/δ f. We would have:

〈 | X1 |2〉 = σ 11 · δ f, 〈| X2 | 2 〉 = σ 22 · δ f (4.39)

where 〈 〉 denotes an average over time, and σ11 and σ 22 are just a convenient notation for
the spectral densities of x1, x2 at frequency f. Note how this equation expresses the idea
that mean power is proportional to bandwidth.
When there is some correlation between x1, x2, there would be some tendency for their
amplitudes of X1, X2 to vary in sympathy, and, although their individual phases would be
quite random, their relative phase would tend to favour some preferred value. This can be
expressed by introducing the cross-spectral density σ12 defined by the relation:

2 = σ 12 · δ f
〈X1 · X*〉 (4.40)

The magnitude of σ 12 is related to the degree of correlation between the variables, and its
phase is the average of the relative phase.
In order to use these ideas to calculate noise temperatures, we need an expression for the
spectral density of a filtered and weighted sum of the two variables. Working in frequency
domain in the normal way, if X is a complex variable expressed as AX1 + BX2, where the
complex numbers A, B are transfer functions of any kind, then the spectral density of X is
given by:

σ = | A |2σ 11 + | B |2σ 22 + 2Re(AB*σ 12) (4.41)

4.5.4 The Dependence on Source Impedance


Armed with the idea of a two-port device as containing two noise generators which will
usually exhibit some degree of correlation, it is straightforward to find the general form of
the dependence of noise temperature on the generator impedance.
The operation of the two-port’s internal noise generators could be observed in at least
three ways: (1) we could short-circuit its terminals, and look at the random currents flowing
at each port; (2) we could leave the terminal’s open circuit, and look at the random voltages
appearing across the terminals; or (3) we could connect infinitely long transmission lines of
specified reference impedances to each of the two ports, and watch the two-port launching
random waves outwards onto these lines. These three approaches correspond to the Y, Z and
S matrix approaches to characterising a two-port.
Rather surprisingly, the first method, based on random currents, will be used here. This is
a definite departure from the usual approach of doing everything with s-parameters. It turns
out to be rather simpler, and most importantly, it leads to the expression for the noise figure
in its most standard form.
The concept of noise temperature is based on referring all the noise generated in the
two-port to a single equivalent noise source in the generator. Using the model described
Figure 4.23, this is done in two steps:
252 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 4.23 Noise model for two-port device

1. in,2 is replaced by an equivalent current generator at the input port.


2. The two-port is now imagined to be noise-free, and the total noise current at the input port
is considered to arise instead from thermal noise in the generator. The noise temperature
Te of the two-port is the physical temperature of the generator impedance required to
do this.

Of course in the real system, the generator might not just be a passive impedance, e.g. it
could be another amplifier, but for the purpose of calculating Te we imagine it to be replaced
by a strictly passive component of the same impedance.
Referring to Figure 4.23, Mason’s rule (see Appendix II) shows that the contribution in,1
makes to the current i2 is given by:

− ZGY21
in,1 × (4.42)
1 + ZGY11

This expression can be made slightly neater by multiplying throughout by YG (= 1/ZG) and
becomes:

−Y21
in,1 × (4.43)
YG + Y11

Hence to replace in,2 by an equivalent current generator at the input, we would have to
multiply it by the reciprocal of this transfer function:
Amplifier Design 253

YG + Y11
in′ ,2 = −in,2 × (4.44)
Y2,1

i′n,2 is the equivalent of in,2 referred to the input side, as shown in the modified flow graph of
Figure 4.23.
The total equivalent current at the input side is now given by:

 Y + Y11 
in′ = in,1 − in,2  G  (4.45)
 Y2,1 

The expression (4.41) can now be used to write down the spectral density of this total
equivalent noise current:

 Y + Y11  * 
2
YG + Y11 
σ = σ 11 + σ 22 − 2Re σ 12 ⋅  G   (4.46)
Y21   Y21  

The object now is just to reduce this to a standard form, treating YG = GG + jGB as the
independent variable and all the other quantities as given constants which describe the
device. To save a good deal of clutter we shall just use symbols for most of the constants
which arise in the working, without bothering to write out their exact expressions.
Inspecting Equation (4.46), the bracket contains no terms in YG higher than quadratic,
and the quadratic term arises only from expanding out the modulus squared term and is
| YG | 2 · σ 22 /| Y21 | 2. Also from expanding this term we get some terms of first order in GG and
jGB, and some constants. The term in Re( . . . ) gives two further terms of first order in
GG and jGB. Grouping together all terms of the same order, the whole expression could
obviously be reduced to:

1
Te = [AG2G + AB2G + HGG + JBG + K] (4.47)
GG

where A, B, H, J, K are suitably chosen constants. Note that the squared terms have the same
coefficient A.
Now let an arbitrary temperature T′ be subtracted from both sides of Equation (4.47), and
on the right-hand side place this term as −T′GG inside the square bracket. The equation
now reads:

1
Te − T′ = [AG2G + AB2G H′GG + JBG + K] (4.48)
GG

where the constant H′ = H − T ′GG.


The method of ‘completing the square’ can now be applied to this equation, rewriting it as
follows:

1
Te − T′ = [A(GG + H′/2A)2 + A(BG + J/2A)2 + K − H′2/4A − J 2/4A] (4.49)
GG
254 Microwave Devices, Circuits and Subsystems for Communications Engineering

Finally, since K − H′2/4A > 0, the constant H′ can be adjusted to make the total constant
term K − H′2/4A − J′2/4A equal to zero. The value of T′ required to do this will be called
Te,min. The expression has now been reduced to:

1
Te − Te,min = [A(GG + H′/2A)2 + A(BG + J/2A)2 ] (4.50)
GG

Finally, the terms H′/2A and J/2A will be rewritten as −GG,opt, −BG,opt respectively, and we
now can write:

1
Te − Te,min = [A(GG − GG,opt)2 + A(BG − BG,opt)2 ] (4.51)
GG

It is now obvious that Te,min is indeed the minimum noise temperature and that YG,opt = GG,opt
+ jBG,opt is the source admittance at which it occurs. We now have:

A
Te = Te,min + [(GG − GG,opt)2 + (BG − BG,opt)2 ] (4.52)
kGG

Finally, re-written in terms of noise figures, this becomes:

Rn
F = Fmin + [(GG − GG,opt)2 + (BG − BG,opt)2 ] (4.53)
GG

where Rn is a new constant. This expression is the one which is useful in practice. Notice
that, as we would expect, it still contains four real parameters, but it has been re-arranged
into a more convenient form. The four parameters: Fmin, Rn, and the real and imaginary parts
of YG,opt.
Note that the parameter Rn, when multiplied by a quantity which has the dimensions of
admittance, gives F which is a pure number. Rn therefore has the dimensions of a resistance,
which is why it was so written. It is called the noise resistance of the two-port. Like the
other noise parameters it is a function of frequency.

4.5.5 Noise Figure Circles

The meaning of the noise figure circles is now obvious. Equation (4.53) for F conforms
to the second class of expression described in Section 4.2.8, and in fact is a special case
where the denominator contains no quadratic terms. The loci of constant F are therefore circles
in the YG plane, and hence also in the ΓG plane, since YG and ΓG are bilinearly related.
The noise figure circles always have the same general appearance. The expression for F
becomes infinite when GG = 0, i.e. for purely reactive generator impedances, which means
all points on the unit circle in the ΓG plane. Thus the circles of constant finite noise figure
cannot cross the unit circle and they must all be nested within it, and shrink down as F falls
onto the optimum point ΓG,opt at which F = Fmin. Their centres all lie on the diameter which
passes through ΓG,opt.
Amplifier Design 255

Noise circles can be drawn for a given device by CAD packages when the four defining
parameters have been supplied. In some cases they are given in device manufacturers’ data
sheets. If really needed, expressions for the centres and radii, though cumbersome, are
straightforward to find by the method described in Appendix I.

4.5.6 Minimum Noise Design

A typical receiving system will consist of a cascade of amplifiers, and possibly other com-
ponents such as filters, mixers, etc. We shall consider the problem of designing the first stage
to minimise the overall noise performance, which can be expressed as:

( F2 − 1)
F(total ) = F1 + (4.54)
GA1

In this case F2 is the total noise figure of all stages except the first, treated as a single unit.
Assuming that F2 is known at least approximately, the result depends only on F1 and GA1,
and these depend only on the ΓG seen by the input of the first stage.
For nearly all transistors and other amplifying devices, the value of ΓG that minimises F
is not the same as the value that maximises GA. This can be seen in Figure 4.24, which gives
specimen noise and gain circles for actual transistors in both unconditionally stable and
potentially unstable regimes. (The figure also illustrates comments made earlier about the
general appearance of gain circles in the two cases.) The optimisation problem is to obtain
the best trade-off between F and GA for the first stage.
Suppose that we choose a trial value of GA for our optimisation. Then the selected ΓG
should be one that minimises the value of F for that value of GA. If we imagine sliding our
ΓG round the chosen GA circle, it will reach a minimum value of F at a point where that
circle is tangent to one of the F circles. Whatever GA is finally chosen, the ΓG should ideally
obey this tangency condition. Thus, as the trial GA is varied, we can generate a correspond-
ing set of ΓG points obeying the condition, and giving a definite best value of F for each GA.
The trial pairs of values of GA and F can then be inserted in the cascading formula to find the
unique ΓG which optimises the overall noise figure.
To perform this optimisation precisely, the noise figure of the second and subsequent
stages needs to be specified. Usually in practice, the gain of the first stage can be made
reasonably high; then the contribution of the second stage will be relatively small, provided
its noise performance is not drastically worse, and the contribution of third and subsequent
stages will be almost negligible. In this condition the optimisation is not very critical, and
it may not be necessary to know F2 very accurately. It will also be found that modest
departures from the tangency condition are not too serious, and that only a few values of
GA will need to be tried when working by hand. (If no estimate of F2 is available in advance,
it may be necessary to assume that all early stages can be made the same and to optimise
the noise measure, as above, of the first stage.)
The optimisation could in any case be done precisely by the optimiser of a CAD package,
but it is a good idea to find a rough solution by inspection first. In principle, every stage in
a receiver chain should be optimised by the same method of trading off gain against noise
figure, but the return from this effort becomes insignificant when the accumulated gain of all
256 Microwave Devices, Circuits and Subsystems for Communications Engineering

previous stages is large. The first stage should be optimised carefully, and some care taken
over the second, but when both these stages have substantial gain, the contribution from
later stages is likely to be insignificant.
Generally a receiver will include a mixer for conversion to a (usually lower) frequency.
A resistive mixer (i.e. one based on non-linear resistive characteristics like normal diode
mixers) will have conversion loss (GA < 1) and will therefore make noise from later stages
more, rather than less, significant. Care should then be taken that sufficient gain is provided
before the mixer stage. (A well-designed resistive mixer has noise figure equal to its conver-
sion loss.) Another important point here is that, with normal mixers, input frequencies both
above and below the local oscillator frequency can produce the same IF (intermediate
frequency). Only one of these frequency bands normally contains a wanted signal, but noise
in both bands can contribute to the IF noise. To avoid a near-doubling of the system noise
level, a filter to pass only the wanted one of the input frequency bands must be placed at
some point before the mixer. If the IF is quite high, the performance required of this filter
need not be very exacting, as it only needs to have substantial rejection at a frequency at a
frequency twice the IF distant from its passband. Precise channel shaping will be left to IF
filters.
A few special receivers use a ‘mixer front end’, meaning that the signal is down-
converted to a low frequency before any amplification. This tends to be done either where an
inexpensive receiver is required and noise is not critical, or at millimetric frequencies where
amplifiers are either extremely expensive or not available at all. In the latter case the mixer
loss must be carefully minimised, and the noise figure of the first IF amplifier is crucial.
In a narrowband design, the input matching network can be designed to give the optimum
ΓG, subject only to manufacturing tolerances and any error in specifying the external source
impedance that feeds into this network. In a broadband design, the optimum ΓG( f ) for the
device will be some function of frequency and it may not be a realistic target to design an
input matching circuit which tracks it precisely.

4.6 Practical Circuit Considerations


4.6.1 High Frequencies Components

It is a common misconception to assume, at high frequencies that all components perform


ideally. In practice, really there is no such thing as an ideal inductor, ideal resistor or ideal
capacitor. All have associated with them either stray inductances, stray capacitances or stray
resistances, meaning that by altering the operation frequency the behaviour of a component
can change quite dramatically. Capacitors may look actually inductive, while inductors may
look like capacitors and resistors may tend to be a little of both.

4.6.1.1 Resistors

The resistor is probably the most commonly used component in electrical engineering.
Everybody is familiar with a resistor. They are used everywhere in circuits, as transistor bias
networks, pads and signal combiner, etc. However, very rarely a thought is given to how
a resistor actually behaves once we depart from DC. In some cases, although the resistor is
Amplifier Design 257

employed to perform a DC function, it can severely affect the RF performance. The equiva-
lent circuit of a resistor at radio frequency is shown in Figure 4.24.

Figure 4.24 Resistor equivalent circuit

R: The resistor value.


L: The lead inductance.
C: The parasitic capacitance. It is the combination of all the internal parasitic capacitances
and can vary depending on the resistor structure.

Bearing this model in mind, it easy to appreciate that the expected impedance performance
is the one shown in Figure 4.25.
As the frequency increases, the effect of the parasitic inductance will also increase, thus
raising the total impedance of the component, until fr where the inductance resonates with the
shunt capacitance. Any further increase in the frequency will decrease the total impedance
since the capacitance will progressively be more significant.

Figure 4.25 Frequency characteristic of a practical resistor


258 Microwave Devices, Circuits and Subsystems for Communications Engineering

There are several types of resistors to choose from but the most important ones are:

1. Carbon composition resistor. They consist of densely packed carbon granules. When
current flows through the resistor they are charged, thus forming a very small capacitor
between each pair of carbon granules. This parasitic reactance influences the behaviour
of the component giving a notoriously poor high frequency performance. A carbon com-
position resistor can experience an impedance drop of more than 70% of the nominal
value even at 100 MHz.
2. Wirewound resistors. They are basically a coil of wire. These resistor types have poor
high frequency performance too. They exhibit widely varying impedances over various
frequencies. This is particularly true of the low resistance values in the frequency range
of 10 MHz to 200 MHz. The inductor L shown in the equivalent circuit is much larger for
a wirewound resistor than for a carbon-composition resistor. Its value can be calculated
in the same manner as for a single layer air core inductor.
3. Metal film resistors. Metal film resistors exhibit the best characteristics over frequency.
The equivalent circuit remains the same, but the actual values of the parasitic elements in
the equivalent circuit decrease.

Figure 4.26 presents the impedance characteristic of the metal film resistor. It can be seen
that with resistor values up to 1 K they can operate within the same limit up to 600 MHz.
The impedance of the metal film resistor tends to decrease especially for high values above
10 MHz. This is due to the shunt capacitance in the equivalent circuit. For very low resistor
values under 50 Ohms the lead inductance and skin effect may become noticeable. The lead
inductance produces a resonance peak and the skin effect decreases the slope of the curve as
it falls off with frequency.

Figure 4.26 Frequency characteristics of metal film and carbon composition resistors
Amplifier Design 259

For high frequency applications thin film chip resistors have been developed. They exhibit
very little parasitic reactance (approximately 1.5 nH) and with care can be used up to 2 GHz.
Typically they are produced on alumina or beryllia substrates.
An alternative at high frequencies is to build up circuits by depositing resistors, cap-
acitors, inductors and interconnectors in the form of metal films on insulating substrates
(MIC and MMIC). Two available technologies exist associated with the thickness of the
metal film:

1. Thick films are usually made by silk screening brushing, dipping or spraying. This pro-
vides a cost-effective method that has been known to operate up to at least 20 GHz.
2. Thin films are made by vacuum, sputtering or chemical deposition. Although more
expensive and complicated it is the only alternative for critical designs and very high
frequencies (can operate above 100 GHz).

Self-assessment Problem
4.13 Using the resistor equivalent circuit shown in Figure 4.24 determine the RF
impedance at 200 MHz and 400 MHz for a 10 KΩ Resistor when:

(a) A 0.5 W metal film resistor having leads of length 21.5 mm and diameter
0.8 mm and stray capacitance of 0.2 pF is used.
(b) A 0.5 W carbon film resistor with leads of 28 mm length and 0.7 mm diameter
and having stray capacitance 0.8 pF is used.

Comment on their comparative performance at both frequencies.

4.6.1.2 Capacitors

Capacitors are used extensively in RF applications such as matching, bypassing, inter-


stage coupling, DC blocks and in resonant circuits and filters. Although in theory an ideal
capacitor can perform equally well in real life, the major task of the designer is to use the
right capacitive structure as well as the right value. The reason for the discrepancy between
the theoretical performance and practice is basically the appearance of additional parasitics
that alter the overall performance. These parasitics have to be taken into account in order to
avoid any surprises in the final design.
The usage of a capacitor is primarily dependent upon the characteristic of its dielectric.
The dielectric characteristics also determine the voltage levels and the temperature extremes
at which the device may be used. Capacitors that are to be used at high frequency must
have low dielectric losses, since as the frequency increases more energy will be dissipated
in the dielectric. This causes a reduction of the overall efficiency or in the extreme even
overheating failure at high and medium power applications.
The equivalent circuit is presented in Figure 4.27.
260 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 4.27 Capacitor equivalent circuit

C: Capacitance.
L: Inductance of leads and plates.
R S: Heat dissipation loss resistance. It models the power transferred to heat due to the
imperfections in the dielectric.
R P: Insulation resistance. It models the power loss due to the leakage through the dielectric.

The effect of the imperfections in capacitor can be seen in Figure 4.28.


Here the impedance characteristic of an ideal capacitor is plotted against that of a practical
one. As the frequency of operation increases, the lead inductance becomes more and
more important. At fr the inductance resonates in series with the capacitance. Above that
frequency, the capacitor behaves as an inductor. Generally larger-value capacitors tend to
exhibit more internal inductance than smaller-value capacitors, due to the larger size of the
plates and leads.
The implications of the internal parasitics can be further clarified if we consider the
1
following. Assume that a capacitor is used in a bypass application, then using Xc = we
ωC

Figure 4.28 Impedance characteristic of a practical capacitor


Amplifier Design 261

can expect that the larger capacitor will result in smaller impedance value and thus better
performance. At high frequencies though the opposite may be true since a larger capacitor
has in general, lower self-resonant frequency. This is something that should be taken into
account when designing an RF circuit at frequency above 100 MHz.
Moreover, since at frequency above fr the capacitor looks inductive, it can be used as one.
When the Q of the capacitor is high and the resistive losses are small, a capacitor operated
in this mode can represent a low-cost high-Q inductor. An additional advantage is the built-
in block by virtue of its capacitance.
The quality factor Q of a capacitor is given by:
Xc
Q= (4.55)
ESR
where ESR is the effective series resistance effectively being the resistance presented by the
capacitor at the specific frequency for which the Q is calculated.
Practical capacitors are usually characterised in terms of the following parameters apart
from the nominal:

• Capacitance tolerance. The maximum change on the nominal capacitive value.


• Temperature range. The range of temperatures in which the capacitor can safely operate.
• Power factor (PF). This is a measure of the imperfections of the capacitor dielectric. It is
given by:
PF = cos φ ⇔ PF = cos (∠ V − ∠ I) (4.56)

• Temperature coefficient. Relates the expected variation in the nominal value for a given
temperature change. Typically is given in ppm/°C.
• Dissipation factor (tan δ). It is the ratio of the effective series resistance at a given
frequency over the reactance of the capacitor at the same frequency:

ESR
tan δ = (4.57)
XC

4.6.1.3 Capacitor types


There are several different dielectric materials used in the fabrication of capacitors, such as
paper, plastic, ceramic, mica. polystyrene, polycarbonate, teflon, oil, glass, porcelain. Each
material has its advantages and disadvantages. The decision of the chosen structure is based
on the application requirements and, of course, the cost.
A table is included that gives a number of capacitor types along with their maximum
usable frequency as well as other important considerations. It is because of the behaviour
shown earlier that sometimes in applications such as wideband decoupling a number of
capacitors of varying value, plus even varying types, have to be used.
4.6.1.3.1 Ceramic capacitor
Ceramic dielectric capacitors vary widely in both dielectric constant (εr = 5 to 10,000)
and temperature characteristics. A rule of thumb is, ‘The higher the dielectric constant εr, the
worse the temperature characteristic.’
262 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 4.29 Radio frequency capacitor types

Low εr ceramic capacitors tend to have linear temperature characteristics. These capacitors
are generally manufactured using both magnesium titanite which has a positive temperature
coefficient and calcium titanite which has a negative temperature coefficient. By combining
the two materials in varying proportions a range of controlled temperature coefficient can be
generated. The capacitors are called temperature-compensating capacitors or NPO (Negative
Positive Zero) ceramics.
They can have temperature coefficients that range anywhere from +150 to −4700 ppm°/C.
Because of their excellent temperature stability, NPO ceramics are well suited for oscillator,
resonant circuits or filter applications.
There are ceramic capacitors available on the market which are specifically designed for
RF applications. Such capacitors are typically high-Q devices with flat ribbon leads or no
leads at all, see Figure 4.29.
The lead material is usually solid silver or silver-plated and contain very low losses. At
VHF these capacitors exhibit very low lead inductance due to the flat ribbon leads. At higher
frequency where lead cannot be tolerated, chip capacitors are used (not lead capacitors).
The common chip capacitor with care can be used up to 2 GHz, though their high internal
inductance of the order of 1.5 nH means care should be taken to avoid the frequency of self
resonance. This point can sometimes make a normally working circuit behave strangely such
as producing a very peaky response. Higher frequency chip capacitor with inductances of
0.3 nH are readily available from companies such as American Technic Ceramics or Dielectric
Lab. Inc.
4.6.1.3.2 Mica capacitors
Mica capacitors typically have a dielectric constant of about 6, which indicates that for a
particular capacitance value, mica capacitors are typically large. The low εr though produces
an extremely good temperature characteristic. Thus mica capacitors are used extensively in
resonant circuits and in filters, when board space is of no concern.
To further increase the stability, silvered mica capacitors are used. In silvered mica
capacitors the ordinary plates of foil are replaced with silver plates. These are applied by a
process called vacuum evaporation which is a very exacting process. This produces even
better stability with very tight and reproducible tolerances of typically +20 ppm/°C over a
range of −60°C to + 89°C, see Figure 4.30.
4.6.1.3.3 Metallised film capacitors
This is a broad category of capacitors encompassing most of the other capacitors not listed
previously. This includes capacitors that use teflon, polystyrene, polycarbonate, etc. As the
name implies, they consist of a thin film of dielectric contained between two thin metalic films.
Amplifier Design 263

Figure 4.30 Low loss interdigitated capacitor

Most of the polycarbonate, polystyrene and teflon styles are available in very tight (±2%)
capacitance tolerances over their entire temperature range. Polystyrene, however, typically
cannot be used above +85°C as it is very temperature-sensitive above this point. (It melts
beyond that point.) Most of the capacitors in this category are typically larger than the
ceramic type equivalent value and are used when space is not a constraint.

Self-assessment Problem
4.14 What is the quality factor of a capacitor with a dissipation factor.
(a) tan δ = 2 10−3;
(b) tan δ = 1 10−4?
Comment on the connection between the quality factor and the dissipation factor.

4.6.1.4 Inductors

4.6.1.4.1 Straight wire inductors


From basic electromagnetic theory we know that even a straight piece of wire will exhibit
some self-inductance at radio frequencies. The value of the inductance associated with a
straight wire is given by:

4J
L = 0.002 J 2.3 log ( − 0.75) (4.58)
d

L: is the inductance in µH.


J: is the length of the wire in cm.
d: is the diameter of the wire in cm.

Inductance becomes more and more important as the frequency increases. At high frequencies
every conductor in a circuit exhibits inductive performance, thus changing the behaviour of
the design.
4.6.1.4.2 Practical inductors
An inductor is usually a wire of some form wound or coiled in such a manner as to increase
the magnetic flux linkage between the turns of the coil. The increased flux through each turn
results in a self-inductance which can reach values well beyond that of the plain wire.
264 Microwave Devices, Circuits and Subsystems for Communications Engineering

Inductors are widespread in radio frequency applications in resonant circuits, filters, phase-
shifters and delay networks, matching networks and as frequency chokes used to control the
flow of radio frequency energy.
The equivalent circuit of a practical inductor is shown in Figure 4.31.

Figure 4.31 Inductor equivalent circuit

Rs: The resistance of the wire.


L: The inductance of the coil.
Cd: The aggregate of the individual parasitic capacitances of the coil.

The physical interpretation is more evident if we consider Figure 4.32. As mentioned


earlier, in order to achieve higher inductance values the proximity of the turns should
increase. That essentially will result in two conductors (here the conducting wound) being
separated by a dielectric material which is a capacitor, assuming there is a voltage drop
between the conductors. Since in real world we always experience wire resistance, (and at
RF this is even more apparent due to the skin effect), a voltage drop will exist between
the windings and Cd will appear as marked in Figure 4.32. The aggregate of all these small
capacitances is called the distributed capacitance.
These parasitics and especially the resistance make the inductor probably the component
that experiences the most drastic changes over frequency.
Based on the equivalent circuit the performance of a practical inductor can be evaluated
with frequency (Figure 4.33). At low frequencies the inductor’s reactance follows the ideal

Figure 4.32 Illustration of parasitics on a practical inductor


Amplifier Design 265

Figure 4.33 Impedance characteristic of inductor

value. This area is basically where we would want the inductor to operate. As the frequency
increases, the reactance begins to deviate from the ideal curve and starts increasing at a
much faster rate. At fr the inductor’s impedance reaches a peak when the inductance
resonates with the parallel capacitance Cd. At still higher frequencies the inductor begins to
look capacitive.
The ratio of the inductor’s reactance to its series resistance is used to measure the
quality of the inductor. The larger the ratio, the better is the inductor. This quality factor is
the Q:

X
Q= (4.59)
Rs

The effect of the resistance therefore is two-fold:

1. The peak at the resonance remains finite.


2. The resonance peak broadens as the resistance increases.

From Equation (4.59) it is evident that the variation of the parasitic elements will result in a
change of the quality factor of the inductor (Figure 4.34).
At low frequency the Q of an inductor increases almost linearly due to the linear increase
of the reactance XL with frequency. As the frequency increases the skin effect becomes more
and more significant and as the resistance increases the slope of the Q gradually decreases.
The flat portion of the curve occurs when the series resistance and the coil reactance change
with the same rate. Above that point, the shunt capacitance and skin effect of the windings
combine to decrease the Q until the resonant frequency of the inductor is reached where it
becomes zero.
266 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 4.34 The Q variation of an inductor with frequency

Figure 4.35 Type of high frequency lumped inductors

The aim of the design of an inductor in most applications is to achieve high Q, compact
size and low capacitance (high self-resonant frequency). In order to design a high Q inductor,
the following rules should be observed:

1. Decrease the AC and DC resistance by increasing the diameter of the wire.


2. Increase the separation of the windings, thus reducing the interwinding capacitance.
3. Increase the permeability of the flux linkage path. This is mostly done by winding the
inductor round a magnetic core material such as iron or ferrite. These types of core,
however, are used primarily for applications below 800 MHz.

At high frequencies, especially in microstrip applications, the inductive values needed are
usually very small (less than 10 nH). It is therefore easier to employ a straight piece of wire
or length on a narrow track, since it becomes impractical to design coil inductors at this
value. In order to make the resulting inductors more compact, designs like the one shown in
Figure 4.35 are employed.
4.6.1.4.3 Single Layer Air Core Inductor Design
In the simplest form a practical inductor consists of a single layer of turns and uses air as the
dielectric, as is shown in Figure 4.36.
A practical formula generally used to design single-layer air core inductors is:

0.394r 2 N 2
L= (4.60)
9r + 10 J
Amplifier Design 267

Figure 4.36 A practical inductor

r: Is the coil radius in cm.


J: The coil length in cm.
N: The number of turns.
L: The inductance in µH.

This formula is accurate within 1% when the coil length is greater than 0.67r. The optimum
Q for such an inductor is achieved when the length of the coil is equal to its diameter (2r).
However, this is not always practical and in many cases the length has to be much larger
than the diameter. In such cases a compromise can be reached when the following solution
are examined:

1. Use a smaller diameter wire, while keeping the length the same, in order to decrease the
interwinding capacitance. However, that will decrease the Q of the inductor since it will
increase the resistance of the wire.
2. Extend the length of coil while retaining the same wire just enough to leave some air gap
between the windings. Nevertheless, the effect again will be that Q will decrease some-
what (since L will decrease) but the capacitance will decrease even more.

Self-assessment Problem
4.15 You are required to design a 10 mm air core having 500 nH inductance. You
can select any one of the following wires (all made from the same material):
(a) a wire with 1 mm diameter;
(b) a wire with 0.5 mm diameter;
(c) a wire with 0.1 mm diameter.
Justify your answer.

4.6.2 Small Signal Amplifier Design


A microwave transistor amplifier can be represented in general by the block diagram of
Figure 4.37.
268 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 4.37 Block diagram of amplifier at radio frequencies

The main blocks of this configuration are:

• Transistor.
• Biasing circuit. The DC biasing circuit determines the quiescent point of the amplifier and
does not contain any microwave components.
• Low pass filter. The LPF is used in order to decouple the microwave circuit from the DC.
• Input matching network.
• Output matching network.

The realisation of the input and output networks determines the performance of the amplifier.
These networks can be called optimising rather than matching networks since the values of
ΓG and ΓL are chosen in order to determine the performance for low-noise or moderate to
high gain. It is important to realise that the I/O matching networks are there not in order
to minimise the input/output reflection coefficient but rather in order to define the behaviour
of the transistor.

4.6.2.1 Low-noise amplifier design using CAD software


The design process of a low noise amplifier is basically the construction of the various sub-
blocks that make up the amplifier. The steps are tabulated in the following:

1. Select a transistor using as guidelines the design specification. The specifications will
include parameters like the noise figure, the gain, the power output, the input and output
reflection coefficients and the price.
2. Calculate the maximum stable gain or available gain and the stability factor within the
band of interest.
3. For unconditionally stable transistors input matching will be determined by selecting a
suitable compromise between the gain and the noise figure. The point that is usually
chosen lies on the tangent point of a constant noise figure and a gain circle.
Amplifier Design 269

4. When the transistor is conditionally stable, plot the regions of instability, by drawing the
input and output stability circles. The matching network is again chosen to compromise
between the gain and the noise figure but additionally the unstable areas must be avoided.
5. Once the input matching network is selected, the output reflection coefficient is obtained
(usually by means of a CAD software) and the output is matched to that value. If k < 1,
then the output stability circle should be drawn to establish that the output reflection
coefficient does not lie in an unstable region.
6. Replace the ideal elements with real-life components. It is up to the designer to include
those parasitic phenomena that make the performance deviate from the ideal case. This
step is critical in order to enable a close approximation of the actual performance. The
design is then optimised using a software tool to determine the optimum configuration.
7. Add the DC biasing and check the stability for the whole frequency range. If the design is
unstable at low frequency, add stability resistors. Fine tune the DC biasing circuit if neces-
sary. If the performance is degraded, fine tune the input/output matching networks as well.

4.6.2.2 Example
Design a LNA to operate between 1.55 GHz–1.80 GHz using the Avantek AT41485 transis-
tor with the following specifications:

Noise Figure < 2 dB


Gain > 12.5 dB

The first step in the actual design is to establish the stability of the amplifier. The input and
output stability circles are therefore drawn so that the behaviour of the amplifier can be
determined. The result is shown in Figure 4.38.

Figure 4.38 Input and output stability circles of the low noise amplifier
in the operational bandwidth
270 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 4.39 Input output unstable areas on compressed Smith chart

This stability check should be performed for a larger frequency range since any existing
harmonics can render the amplifier unstable. However, the introduction of a biasing network
in the input and output introduces a resistive load that stabilises the amplifier at low frequency.
Only an initial check is performed at this stage, since a more vigorous analysis is due at later
stages.
Once the unstable areas are determined, as shown in Figure 4.39, the operation point is
chosen so that these are avoided. The operation point is selected as a compromise between
two antagonistic parameters noise figure (NF) and the available gain (GA). If we deter-
mine the optimum noise figure (NFo) and the maximum available gain Gmax, then it is easy
to realise that these are two separate points on the ΓG plane. A compromise, therefore, is
essential in order to achieve the optimum performance which is application-oriented. In
Figure 4.40 the noise figure circles and the gain circles are presented.
The adjoining point of any two circles results in the optimum performance for both para-
meters. Which set of circles is used depends on the designer and the application. Attention
should be drawn though to the fact that a margin should be allowed for manufacturing
degradation. Thus, a choice at this point of a gain of 12.5 dB (in our example) or close
by will certainly result in a circuit that does not satisfy this condition. Allowing for this,
therefore, the selected values are presented in Figure 4.41.
Because the plane on which the calculation is performed is the ΓG plane, any reflection
coefficient calculated by the CAD software will be a source reflection coefficient. If,
therefore, the normal matching procedure is to be followed, the conjugate of ΓG should be
used.
Once the required source reflection coefficient ΓG is determined, the output reflection
coefficient Γout can be calculated from this and the s-parameters of the transistor, using:
Amplifier Design 271

Figure 4.40 Constant noise figure and gain circles at 1.675 GHz

Figure 4.41 Selected noise figure and gain for the low noise amplifier design

s12 ⋅ s21 ⋅ ΓG
Γout = s22 + (4.61)
1 − s11 ⋅ ΓG

or the appropriate CAD command (which performs the same calculation). When these
two reflection coefficients are established, the input and output matching networks are
designed.
272 Microwave Devices, Circuits and Subsystems for Communications Engineering

4.6.3 Design of DC Biasing Circuit for Microwave Bipolar Transistors


The biasing for high frequency circuits, although essential, has often been overlooked.
The selection of the quiescent point heavily influences the characteristics of the amplifier
Essentially high gain, high power, high efficiency and low-noise are directly influenced by
the biasing. Furthermore, hasty bias design can render the amplifier unstable.
The aim of a good DC bias design is:

• to select the appropriate quiescent point;


• to keep it constant over variations in transistor parameters and temperature.

The selection of the DC quiescent point for an HF transistor depends very much on the
application. Thus, the most common selections are presented in Figure 4.42.

Figure 4.42 Quiescent points for bipolar transistors

A → Low noise and low power.


B → Low noise and high gain.
C → High power in class A.
D → High power in class AB or B.

Depending on the components used, biasing circuits are divided in two categories, passive
biasing circuit (purely resistive) and active biasing circuits. Resistor bias networks can be
used with good results over moderate temperature changes. However, active bias network
are necessary when large temperature changes are encountered.

4.6.3.1 Passive biasing circuits


The two most common configurations are presented in Figure 4.43:
Amplifier Design 273

Figure 4.43 Passive biasing circuits for bipolar transistors

These two configurations are often used at microwave frequencies. The voltage feedback
with constant base current circuit produces smaller resistive values which makes it more
compatible with thin or thick film technology. In order to improve stability a bypass resistor
is added at the emitter of the transistor. This method, however, is useful only at the low
range of microwave frequencies. A more detailed analysis of the two circuits is given in the
following paragraphs.
4.6.3.1.1 Voltage feedback biasing circuit
The starting point is as always the selection of the quiescent point. Once this is chosen, the
typical value for the forward current gain (β ), and the base-emitter junction voltage (VBE) are
determined from the transistor data sheets. The following set of equations is then used to
calculate the values of the resistors:

IC
IB = (4.62)
β

VCE − VBE
RB = (4.63)
IB

VCC − VCE
RC = (4.64)
IC + IB
274 Microwave Devices, Circuits and Subsystems for Communications Engineering

The last parameter needed is the value of the voltage source VCC. This parameter is selected
by the designer so that adequate power is provided to the design. A value larger than VCE
should be selected but not much larger since this reduces the efficiency of the amplifier.

4.6.3.1.2 Voltage feedback with constant base current


Again the analysis starts with the selection of the quiescent point from the specifications for
the scattering parameters and the noise figure. The values of β and VBE are then determined.
The current at the base of the transistor is:

IC
IB = (4.65)
β
Also using Kirchhoff’s Laws we have:

VBB − VBE
RB = (4.66)
IB

VCE − VBB
RB1 = (4.67)
IB + IBB

VBB
RB2 = (4.68)
IBB

VCC − VCE
RC = (4.69)
IC + IB + IBB

For the correct operation of the biasing network it is essential that IB Ⰶ IBB. This enables
the biasing circuit to retain a stable quiescent point. In order to certify this a relation is
chosen between the current in these branches. A satisfactory value is:

IBB ≈ 10 · IB (4.70)

Even with this assumption it is evident that the number of unknowns still exceeds the
number of equations. Thus, an appropriate value is selected for one parameter. Usually
the choice is between RB or VBB. It is recommended that RB should be a few KΩhms in
order to establish that IB is quite small. Usually a practically available value is chosen,
e.g. 22 KΩhms.
Thus, after determining VBB from

VBB = VBE + IB RB (4.71)

we can substitute in Equations (4.61) to (4.63) to determine the other resistors.

4.6.3.2 Active biasing circuits


When large temperature variations are expected and close control of the quiescent point is
required, usually an active biasing circuit is employed. The active biasing circuit is actually
a feedback loop that secures the collector current of the microwave transistor and adjusts the
Amplifier Design 275

Figure 4.44 Active biasing circuit for bipolar transistors

base current to hold the collector current constant. For improved temperature performance
the two devices should be situated as close as possible, in order to experience the same
temperature variations.
A commonly used biasing network is presented in Figure 4.44. A P-N-P BJT (Q1) is used
to stabilise the operating point of the microwave transistor (Q2) over any variations of the
quiescent point. The actual value of the quiescent point can be regulated by Rb2 and RE, RB2
controls the voltage VCE and RE controls the collector current IC.
The operation of the network is qualitatively explained as follows. If IC2 tends to increase,
the current I3 increases and VEB1 decreases. The reduction of VEB1 decreases IE1, which in turn
decreases IC1 and IB2. Thus, it follows that the tendency of IC2 to increase is combated from
the biasing network, therefore producing a closer bias stability.
The mathematical analysis for the circuit of Figure 4.44 is presented next:

The base current IB2 is:

IC2
I B2 = (4.72)
β2
Assuming IC1 Ⰷ IB2 we can give a value to IC1 such that the inequality is enforced.

IC1 ≈ 10 · IB2 (4.73)

VBE
Therefore: RC = (4.74)
IC1
276 Microwave Devices, Circuits and Subsystems for Communications Engineering

IC1
Also for Q1: IB1 = (4.75)
β1

(β1 + 1)
IE1 = IC1 (4.76)
β1

At the emitter of Q1: I3 = IC2 + IE1 (4.77)

VCC − VCE2
and therefore: RE = (4.78)
I3

As earlier we should satisfy I1 Ⰷ IB1 and therefore:

I1 ≈ 10 · IB1 (4.79)

VEB1 + I3 RE
And also: RB1 = (4.80)
I1

VCC − I1 RB1 
Finally: RB2 =  VCC − I1 RB1
I2  ⇒ RB2 = (4.81)
I1
But I1 Ⰷ IB1 ⇒ I1 ≈ I2 

Note: The assumption that the base current is much smaller than the vertical branch is a
general one and has to be satisfied to extract the optimum performance. The choice of the
base current being an order of a magnitude smaller is just an approximation that usually
yields good results. It is up to the designer to decide whether it is adequate for the particular
application or a more stringent one is required.
From the standpoint of temperature compensation, passive biasing circuits are inexpensive
to build and can provide satisfactory results. However, when the DC current gain variation is
large, automatic compensation can be achieved only by an active biasing circuit. Furthermore,
the active biasing circuit can provide better operating point stability, especially for low noise
or high power amplifier. The major drawbacks to the active biasing circuit though are added
complexity and cost.

Self-assessment Problem
4.16 For a bipolar transistor having VBE = 0.7 and β varying between 30 (min) and 300
(max) and typical value 150, design:
(a) A voltage feedback biasing circuit.
(b) A voltage feedback with constant base current biasing circuit.
(c) An active biasing circuit.
for a quiescent point of VCE = 5V and IC = 6 mA. The voltage supply is VCC = 20 V.
Amplifier Design 277

Figure 4.45 Quiescent points for GaAs FET

4.6.4 Design of Biasing Circuits for GaAs FET Transistors


The general aspects for the design of the biasing circuit are valid for both GaAs and silicon
transistors. There are differences in the actual layouts since the two transistor types have
different characteristics. The selection of the quiescent point in a GaAs FET is again applica-
tion dependent. Figure 4.45 shows typical GaAs FET characteristics and biasing points.

(A) → Low power low noise operation


It operates at a relative low drain source voltage (Vds) and current (Ids), with
Ids typically equal to 0.15 Idss. The type of operation is usual in Class A mode
(low noise).
(B) → High gain low noise
The drain voltage remains low but the drain current is increased to 0.90 Idss for
high power gain.
(C) → High power high efficiency. Class A operation
To achieve higher power output the drain voltage Vds must also be increased. In
order to maintain linear operation in Class A the drain current must be decreased.
The recommended values are Vds ≈ 8 − 10 V, Ids ≈ 0.5 Idss.
(D) → High power high efficiency. Class AB or C operation
For higher efficiency or to operate in Class AB or B the drain to source current
is decreased further.

The most typical passive designs for GaAs FET amplifiers are presented next.

4.6.4.1 Passive biasing circuits


1. Bipolar power supply
This DC biasing circuit requires two power supplies. When the source is connected directly
to the ground terminal the source inductance can be made relatively small. By doing so low
noise high gain, high power and high efficiency can be achieved at higher frequency.
278 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 4.46 Bipolar power supply circuit

2. One sign power supply


To remove the requirement for two sources with opposite signs, the following two configu-
rations are used (Figure 4.47):

(a) Positive Supply Circuit (b) Negative Supply Circuit

Figure 4.47 Single power supply biasing circuits

These types of biasing circuits need very good microwave bypass capacitors at the source.
This may cause a problem at higher frequency because any small series source impedance
could cause high noise and possible oscillations.
Note: The previously mentioned configurations require caution during the start-up phase
in order to prevent the burnout of the transistor. The proper start-up procedure is to establish
first that a negative voltage is applied at the gate before the drain is connected to the supply.
To avoid these difficulties delay circuit is included to certify that the Vgs is applied before Vds
or a Unipolar power source is employed (Type 3).

3. Unipolar power supply


These two DC biasing circuits require only one power source. They employ a source resistor
to provide the turn-on turn-off transient protection. However, the efficiency and noise figure
will be degraded, since the source resistor will dissipate some power and generate noise.
Additionally low frequency oscillations can appear due to the source bypass capacitor.
Amplifier Design 279

VS = IdsRS < + VDS VS = −Ids RS < − Vgs


(a) Unipolar Positive Supply Circuit (b) Unipolar Negative Supply Circuit

Figure 4.48 Unipolar power supply biasing circuits

Figure 4.49 GaAs biasing circuit with Zener diodes

The value of RS is adjusted to provide the right VS for a proper quiescent point and
transient protection.
For further protection it is common to shunt the decoupling capacitors with zener diodes.
The diodes provide additional protection against transients, reversing biasing and over-
voltage. The overall configuration is shown in Figure 4.49.

4.6.4.2 Active biasing circuits


Besides the aforementioned passive DC biasing circuits, when large parameter variations are
expected it is usual to employ active circuits like the one in Figure 4.50.
The operation of the active network is similar to the silicon transistor case. The quiescent
point is again controlled by RB2 and RE . RB2 is adjusted to provide the required Vds and RE
controls the drain current.

4.6.5 Introduction of the Biasing Circuit


A very important aspect of the implementation of the biasing circuit is introducing the
biasing (DC power) to the radio frequency network. The application of the bias should be
280 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 4.50 Active biasing circuit for GaAs transistors

performed in such a way that the performance of the amplifier is not affected over the
intended operating band. There are two objectives in introducing the DC.

1. Minimise the amount of radio frequency power that is lost down the bias lines since this
will obviously reduce the amplifier’s performance (noise gain).
2. The impedance presented to the main input and output radio frequency lines, by the bias
lines, should not affect the matching arrangements at the operating frequency.

The coupling of the DC and RF parts of the amplifier is achieved by the use of components
like inductors and capacitors. The inductors usually known as radio frequency chokes (RFC)
effectively operate as blocks for the high frequency signals, thus controlling the flow of RF
power to the chosen path. Similarly the capacitors are used as DC blocks thus directing the
DC power to the required nodes. With appropriate use of these components it is possible to
design circuits that achieve very high efficiencies by maximising the DC to RF conversion
power.
The use of ideal components for decoupling is illustrated in the following designs.

1. Common base
The requirement for a common base for the RF transistor makes the ground node different
for the RF and the DC circuit. The solution is to employ inductors and capacitors to separate
the two circuits as shown next (Figure 4.51).

• The capacitance Cb is used to block the DC grounding, but permitting the AC ground.
• The RFCE is employed in order to ground the DC signal without presenting a ground for
the AC signal.
• RFCC, RFCB are used to prevent the RF power from flowing in the biasing circuit.
Amplifier Design 281

VdC

RC

RFC
r.f. INP r.f. OUT

RFC
RFC
RB
CB

Figure 4.51 Common base configuration

2. Common collector
As in the previous case the necessary circuit is given in Figure 4.52.

VdC

RC
RB

RFC r.f. OUT


r.f. INP

RFC RFC

CC

Figure 4.52 Common collector configuration


282 Microwave Devices, Circuits and Subsystems for Communications Engineering

d.c. bias

Figure 4.53 Introduction of biasing using practical inductors

4.6.5.1 Implementation of the RFC in the bias network


At the frequency range that most RF amplifiers operate, we cannot assume that the com-
ponents are ideal. Therefore the ideal inductor employed earlier to provide the isolation
between the DC and AC circuits is the obvious choice. Other techniques that can perform
adequately even at high frequencies are presented next.

1. Inductor
In the simplest case when the operational frequency is sufficiently low (less the 600 MHz)
then commercially available inductors can be used, i.e. SC30, SC10. The inductor is simply
connected to the appropriate lead as is shown in Figure 4.53.
As the frequency increases the construction of simple inductors to operate well enough
becomes progressively more difficult. For higher frequencies therefore it is common to use
other techniques to provide the isolation between.

2. High impedance line


At high frequencies a length of a low impedance line is exhibiting inductive performance.
Using this phenomenon we can employ a thin line as an inductor (Figure 4.54). The use
of the high impedance line will allow the introduction of DC bias without affecting radio
frequency performance significantly. Manufacturing limitation on the width of the line make
this method difficult to implement if the frequency increases further.

3. Quarter wavelength shorted stub


From line transmission theory it is clear that a quarter wavelength line is basically a normalised
impedance inverter, that is the normalised impedance looking in the input is the inverse of
the output normalised impedance. If therefore a short-circuited quarter wavelength line is
used, as shown in Figure 4.55, the impedance loading the RF circuit at point A will be
infinite which means that there will be no RF power leakage to the biasing network.
To prevent the short circuit appearing at the biasing network a capacitor is used in
series with the ground. The short circuit/capacitor combination can be implemented as a low
impedance line (Figure 4.57a) acting as a parallel plate capacitor to the ground or a lumped
Amplifier Design 283

d.c. bias

Figure 4.54 Introduction of biasing using a thin transmission line

λg
4

Figure 4.55 Quarter wavelength line configuration

bypass capacitor (Figure 4.56). If more wideband performance is required, then a radial stub
can be used (Figure 4.57b). The gradual change of the width of the termination gives better
performance over a larger frequency range.
When a distributed element is used for the construction of the short circuit, the dimensions
of the patch should be selected in order to prevent the appearance of standing waves. The
width of the short circuit therefore should not exceed half a wavelength (λ g/2), to prevent
the patch from becoming a microstrip antenna.
284 Microwave Devices, Circuits and Subsystems for Communications Engineering

d.c. bias

λ
4

Figure 4.56 Introduction of biasing using short circuited λ /4 line and lumped capacitor

d.c. bias
λg
ᐉ << 2
λg
d.c. bias ᐉ << 2

λg
4 λ
4

(a) (b)

Figure 4.57 Introduction of biasing using short circuited λ /4 line and distributed capacitor
Amplifier Design 285

d.c. bias

λg
j << 2

λg
4

Figure 4.58 Introduction of biasing using thin λ /4 line and distributed capacitor as short circuit

An improvement of the high impedance line and λ /4 shorted stub can be achieved when
λ /4 high impedance line is used (Figure 4.58). This is the most common technique in micro-
wave design since it combines the advantages of both methods. The impedance of the line
does not have to be very small (around 0.5 mm) which is well within most manufacturing
procedures’ capabilities. When short circuit stubs are used for input/output matching, the
biasing can be introduced directly to them, instead of introducing additional stubs. This
method should be employed with more care since any design errors in the short circuit
would seriously affect the matching and subsequently the amplifier performance.

4. Filter network
For designs where the isolation is critical, low pass filters can be employed. This naturally
means that the cost of the design will be much higher. The implementation of the filter is
dependent on the operational frequency and bandwidth of the design. The general layout of
the filter is presented in Figure 4.59.

L L

Bias C C LPF

Figure 4.59 LPF network


286 Microwave Devices, Circuits and Subsystems for Communications Engineering

d.c. bias
C

Figure 4.60 LPF using distributed elements

d.c. bias

Chip capacitor
C

Wire bonds

Figure 4.61 Semi-lumped LPF configuration

This can be implemented using either distributed or lumped components. In Figure 4.60
the design of the LPF is implemented using microstrip lines. Although easier to include in
a microstrip board, the use of distributed elements can be limiting in their operating band-
width and can cause spurious responses at higher frequencies. For multi-octave circuits it is
necessary to use the semi-lumped filter arrangement shown in Figure 4.61.
Amplifier Design 287

Self-assessment Problem
4.17 What network would you use to introduce the biasing circuit when a narrowband
amplifie’s design is when:
(a) the centre frequency is 4 GHz;
(b) the centre frequency is 500 GHz?
Justify your choice.

4.6.5.2 Low frequency stability


Because the microwave transistors are potentially unstable at low frequency it is extremely
important to ensure low frequency stability. Otherwise the appearance of any harmonic at
low frequency can render the amplifier unstable.
The resistive loading from the biasing circuit is enough to extend the stable region to a few
hundred MHz. In order to further ensure the stability of the amplifier it is common to include
two additional load resistors so that they appear at low frequency as input, output loads.
The two available ways of configuring these resistors are presented in Figure 4.62.
Practical values for these resistors vary from 20–50 Ohms. In extreme cases even higher
values can be included.
Note: In the design of Figure 4.57(b) the stability resistors are affecting the biasing
circuit since they are connected in series to the base and collector resistors. It is therefore
important that their values are included in the calculations. In the other configuration the
stability resistors are not part of the biasing since they are grounded through a capacitor,
which removes them from the DC equivalent circuit.

VCC
RC

RC RS2
RB1 C
RB1

RS2

CS2 RB RS1
RB

RS1 RB2 CS1


RB2
CS1

(a) (b)

Figure 4.62 Configuration of stability resistors


288 Microwave Devices, Circuits and Subsystems for Communications Engineering

4.6.5.3 Source grounding techniques

A major problem in the manufacturing of an RF amplifier is to find an effective method of


grounding the emitter (or source in the case of GaAs transistors) without the introduction
of significant parasitic inductance. From the earlier discussion it should be appreciated that
any emitter inductance will result in series feedback tending to reduce the gain degrade the
noise figure and potentially destabilise the amplifier.
Several possibilities for grounding the emitter are presented next. The common charac-
teristic in all of them is the effort to minimise the length and/or maximise the diameter of the
emitter lead.

1. Topside ground plane, tapered in to meet the source leads


Given the very small dimensions of the device and the possibility of coupling between the
main radio frequency lines and topside ground plane, relatively long thin sections of line
would be required to meet the source leads (Figure 4.63). A significant inductance would
therefore result in making this a viable method only in a lower GHz region.

2. Plated through holes


In this method holes are drilled through to the ground plate, with the walls of the holes
being metallised (Figure 4.64). This is an attractive option for use up to 10 GHz. However,
it requires a more complicated manufacturing process for producing plated through holes,
therefore being significantly more expensive.

3. Raised ground plane


This method introduces the least inductance. The device simply sits on the ground plane
with the emitter (source) soldered to it (Figure 4.65). However, given the close proximity
of the bias lines to the device, coupling between the ground plane and the bias lines would
inevitably affect the low-pass performance of the biasing network. It is possible to reduce
the amount of ground plane that is raised to lessen this problem. Additionally it necessitates
specific preparation of the PCB which results in increased cost.

Ground
Printed
Circuit

Substrate

Figure 4.63 Topside ground plane


Amplifier Design 289

Figure 4.64 Plated through hole

Figure 4.65 Raised ground plane

4. Metallic stud through holes


Metallic stud through holes are drilled in the board directly under the source leads. To
achieve effective electrical grounding two bias studs are flush fitted into the holes in the
main brass base, separated by the width of the device package. Holes, the same distance
apart, are drilled through the substrate, directly below the position of the emitter (source)
leads and as close to the package of the device as possible. The substrate is then firmly fixed
to the main base using highly conductive silver loaded epoxy. The use of screws to hold
290 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 4.66 Metallic stud through holes

down the substrate should be avoided, since the resulting localised pressure has the tendency
to bend the substrate. This can lead to air gaps between the substrate ground plane and the
brass base and can give spurious transmission effects.

4.7 Computer Aided Design (CAD)


It took little more than ten years for the highly expensive software that was once only avail-
able on mainframe computers to develop in a relatively inexpensive tool that all engineers
can have on their personal computer. In the early days Computer Aided Design was the
prerogative of large companies that could afford the cost of both software and hardware.
Thanks to the rapid pace of development in the computer hardware and software technology
we are now at a stage when CAD tools are available even as freeware.
To understand the how the CAD approach revolutionised RF design a look into the early
days of RF design is needed. Figure 4.67 illustrates the classical procedure for the design of
microwave circuits. The first step is to identify the specification for the required application,
using them to derive an initial circuit configuration. Available design data and previous
experience are really the only tools in the first steps of the design. The circuit parameters
are then determined, employing a series of analysis and synthesis techniques. A laboratory
prototype is then constructed and its performance is measured and compared with the desired
specifications. If the specifications are not met – which is the most probable outcome – the
circuit is modified. Adjustment, tuning, trimming mechanisms are used in order to modify
the obtained performance. Measurements are continuously carried out until the required
specifications are met. The final configuration is then fabricated.
The above procedure had been in use for numerous years before the advent of computer
technology. However, it was evident that as the technology of RF circuits was advancing,
Amplifier Design 291

Circuit Design
Specifications

Initial Cicuit Design


Design Data

Generation of
Laboratory Modifications
Prototype

Fail

Measurement of
Compare
Prototype
with specifications?
performance

Pass

Finalised
Design

Figure 4.67 Classical RF circuit design approach

the application of this iterative and quite empirical method was becoming increasingly more
difficult and costly, due to the following considerations:

• The increasing complexity of the modern RF circuits makes the analysis and the unaided
design of the initial circuit if not impossible more time-consuming.
• The required specifications are usually more stringent, thus making the effect of tolerances
more and more important.
• The huge variety of components that are currently in existence makes the choice of the
appropriate device a very tedious task.
• The use of MIC and MMIC technology for the fabrication of most RF circuits makes any
modifications to the prototype circuit very difficult, thus making prohibitive the cost of
the iterative process.

4.7.1 The RF CAD Approach


To provide a solution to these problems computerised tools have been developed that enabled
the prediction of the performance of the initial prototype without actually manufacturing it.
292 Microwave Devices, Circuits and Subsystems for Communications Engineering

Circuit Design
Specifications

Design Initial Cicuit


Data Design

Component
Schematic
and Layout
Capture
Libraries

Analysis
Optimisation
of Design

Fail

Compare
with specifications?

Pass

Fabrication of
Laboratory Modifications
Prototype

Fail

Measurement
Compare
of Prototype
with specifications?
Performance

Pass

Final
Circuit
Design

Figure 4.68 CAD RF circuit design procedure


Amplifier Design 293

In essence, by Computer Aided design it is implied that a number of the previous steps are
now performed with the use of the computer thus making the whole process more reliable,
more accurate, cheaper, less time-consuming and more comprehensive.
The amended design process that incorporates the computer as a tool is depicted in
Figure 4.68. Effectively the process consists of two nested loops. In the inner one the com-
puter is employed to evaluate the performance of the initial circuit. Numerical models for
the various components are recalled from the libraries of subroutines, when needed in the
analysis, now totally performed by the computer software. The estimated performance is
then compared with the desired specifications and if these are not met, the circuit parameters
are altered in a systematic manner. Several strategies exist that direct the manner in which
the change occurs towards the optimum performance. The sequence of circuit analysis com-
parison with the desired performance and parameter modification is performed iteratively
until the specifications are met or the best attainable performance (within the given constraints)
is reached. The circuit is then fabricated and experimental measurements are obtained.
Additional modifications may still be necessary if the modelling and/or analysis has not been
sufficiently accurate. However, these if needed, are usually small, thus minimising the number
of experimental iterations.
The process of CAD design as outlined comprises of three major segments:

• modelling
• analysis
• optimisation.

In order to appreciate the benefits and limitations of the CAD design approach, let’s examine
each of these segments in more detail.

4.7.2 Modelling
Modelling basically involves the characterisation of various active and passive components
to the extent of providing a numerical model that can be analysed by the computer. This
implies that a model exists in a pre-designed database that contains information about a
variety of components that can be employed in an RF circuit. The capabilities of any soft-
ware tool can be limited or enhanced depending on the extent of that library. Commercial
packages nowadays contain data on any number of elements as shown in Table 4.3.
The list in Table 4.3 is by no means complete, it is really just an indication of available
elements, each of which can contain hundreds of components. The extent to which one can
take this list of RF elements evidences the difficulty of modelling. All these structures need
to be characterised fully in terms of impedance, phase velocity losses, etc. Additionally, it is
vital to model parasitic reactances as well, in order to obtain an accurate evaluation of the
actual performance.
Difficulties in modelling have hindered the use of CAD techniques at RF frequencies.
There is the need to identify simplified equivalent circuits and closed form expressions
that possess sufficient accuracy for circuit design. Typical examples of this are microstrip
discontinuities such as the T-junction shown in Figure 4.69. Quasi-analytical models published
in the literature have been included in RF CAD tools to simulate this effect on microstrip
circuits. So whenever a discontinuity effect is to be simulated, the equivalent representation
294 Microwave Devices, Circuits and Subsystems for Communications Engineering

Table 4.3 Models found in commercial packages

Element Features

Semiconductor devices BJT


MOSFET
MESFET
HEMT
Point contact
Schottky barrier detectors
Varactor
PIN diodes

Passive elements Electron and avalanche devices


Lumped elements
Coaxial cables
Microstrip lines
Finlines
Striplines
Waveguides
Coplanar waveguide lines
Slot lines
Lumped components
YIG and DRO resonators
Planar circuit elements

General RF elements Air core inductors


Chip capacitors
Chip resistors
Piezoelectric crystals
Transmission line transformers
Resistive pads
Balun transformers
Toroidal ferrite-core inductors

Fabrication specific elements MMIC elements


Surface mount technology elements

Signal sources AC and DC voltage and current sources


AC power sources
Periodic waveform sources (sawtooth square wave, etc.)
Noise sources

is introduced in the schematic whereupon the simulator will substitute for the analysis the
equivalent electrical circuit.
For example, when a vertical connection between two transmission lines (very common
in stub matching networks) is needed, the T-junction has to be introduced in the design in
order to allow for the discrepancies on the Microstrip field. Figure 4.70 shows a schematic
diagram of the configuration. In reality, whenever the representation shown in Figure 4.69a
Amplifier Design 295

Figure 4.69 Microstrip T-junction in Libra Series IV

Figure 4.70 Schematic caption of single stub matching network


296 Microwave Devices, Circuits and Subsystems for Communications Engineering

is encountered in a design, it is effectively substituted with the equivalent circuit shown in


Figure 4.69c (the calculation of the parameters is performed using a set of semi-empirical
closed expressions inherent in the model).
Modelling is a vital part of evaluation process. In reality, the error in the performance
evaluation of a circuit is at least as large as the modelling error. It is essential therefore
to understand the models employed as well as the way that these are introduced. If in
Figure 4.69 the T junction is not inserted, then the actual performance will differ from the
expected. Similarly if the limits of the model (e.g. if w1 < 0.1H) are exceeded, the actual
results can vary significantly.

4.7.3 Analysis

Analysis provides the response of a specified circuit configuration to a given set of inputs.
This is probably the most developed and widely accepted aspect of CAD. Microwave
circuit analysis usually involves evaluation of s-parameters of the overall circuit in terms of
the given s-parameters of the constituent components.
Modern RF CAD tools contain several different strategies to allow the designer to analyse
different circuits fast and accurately. The need for different techniques arises from the variety
of issues involved in analysing complex RF circuits. It is self-evident that different types of
problems require different approaches. That applies not only in terms of the required outcome
(e.g. a different approach is needed when the input reflection coefficient and the third-order
intercept point of a power amplifier are evaluated), but in terms of the actual circuit design
(a low noise amplifier and an oscillator require different analysis techniques). The basic
amplifier design can be carried out using only linear analysis based on the s-parameter matrix
or small signal equivalent model. However, in applications where small signal assumption is
not satisfied, such as mixers, oscillators or even power amplifiers, then a nonlinear simulation
is needed.
The most common linear and nonlinear analysis techniques are summarised next.

4.7.3.1 Linear frequency domain analysis


RF circuit analysis is usually performed in frequency domain. Working in the frequency
domain implies that only the steady state response of distributed elements is considered.
This is significant in microwave circuits where the multiple reflections encountered and the
frequency dependence of the line parameters can be very difficult to analyse in the time
domain.
Linear frequency domain analysis results in a time-independent solution for linear net-
works with sinusoidal excitation. Nonlinear components are analysed using small-signal
models. Y-parameters matrix techniques are used to solve for the steady state response,
treating individual components as frequency dependent admittances within a nodal matrix.
Matrix reduction techniques are employed to determine the Y-parameters of the overall
circuit.
Linear analysis can perform noise measurements considering temperature-dependent
thermal noise from lossy passive elements as well as temperature-dependent and bias-
dependent thermal noise for nonlinear devices.
Amplifier Design 297

4.7.3.2 Non-linear time domain transient analysis


Non-linear components like transistors are contained in the equivalent circuits, amplitude-
dependent. This necessitates the use of time domain analysis in order to obtain their true
response. Spice-type transient analyses are commonly used in such cases. This involves the
solution of a set of integro-differential equations that express the time dependence of the
currents and voltages of the analysed circuit.
The need to consider losses, dispersion for distributed elements, and parasitics make the
use of time domain simulators inefficient especially at high frequencies. The difficulty arises
from the fact that in the time domain all of the RF elements are represented by simplified
frequency independent models.

4.7.3.3 Non-linear convolution analysis


Convolution analysis is a time domain analysis strategy that circumvents the inefficiencies by
modelling the frequency dependent elements in the frequency domain. The frequency domain
representation is transformed to the time domain effectively giving the impulse response of
the elements. This is subsequently convolved with input signal to provide the output. To reduce
the computation time, elements with exact equivalent circuits are modelled in the time domain.

4.7.3.4 Harmonic balance analysis


The harmonic balance is an iterative process treating a nonlinear circuit as a two-part net-
work comprising of a linear sub-network and a nonlinear element (Figure 4.71). The voltages
and currents flowing from the interface into the linear part of the circuit are determined
by using frequency domain linear analysis. Currents and voltages into the nonlinear part
are calculated in the time-domain. Fourier analysis is used to transform the time domain to
the frequency domain. Kirchhoff’s Current Law states the currents should sum to zero at the
interface. Any error that is encountered is reduced by successive iterations. When the method
converges, the currents and voltages approximate the steady-state solution
The large number of calculations necessitated by each iteration combined with the
requirement for several iterations before convergence imply that a fast processor and a lot
of memory is needed. The strain on the computer resources increases as the number of
harmonics increases. For example, as much as 300 Mbytes of RAM are needed to handle
three independent frequencies with 12 harmonics.

Nonlinear
Linear Network
Device

Frequency domain Time domain

Figure 4.71 Representation of nonlinear circuit for harmonic balance analysis


298 Microwave Devices, Circuits and Subsystems for Communications Engineering

4.7.3.5 Electromagnetic analysis


A different form of analysis often used at RF and microwave frequencies is electromagnetic
analysis. Electromagnetic simulators basically solve Maxwell’s equations for the analysed
circuit in two and three dimensions. The electromagnetic simulators should be able to cope
with the general metal-dielectric structure problem. To simplify the general case the trans-
mission lines (metal structure) can be assumed to be infinitely thin. This assumption reduces
the required computation time by an order of magnitude.
In the former case, a three-dimensional type analysis is used to determine the field
patterns surrounding a metal structure embedded in various layers of dielectric materials.
In the latter case only a two-dimensional analysis is needed to calculate the modal character-
istics and electromagnetic fields for a cross-section of transmission lines embedded between
layers of dielectric materials. The calculated characteristics can be impedances, voltages,
currents, powers, propagation velocities and effective dielectric constants.

4.7.3.6 Planar electromagnetic simulation


This is usually used to solve for the s-parameters of arbitrary microstrip structures. By limiting
the problem in two dimensions the necessary processing time is dramatically reduced. Some
of the most recent tools can handle multi-layered structures with vias and varying dielectric
thickness. Nevertheless the structures are still assumed to be planar and so the term 21/2-D is
used (two-dimensional currents but 3-dimensional fields). The speed of planar electromagnetic
simulators makes them a very attractive solution when non-standard structures are used.
4.7.3.6.1 3-D electromagnetic simulation
These simulators can analyse arbitrary-shaped metal structures positioned over multi-layered
dielectric structures, in an enclosed metal shielded environment. They use finite element
techniques to divide the problem space into a point mesh and calculate the field intensity at
the vertices. Typically they are employed when analysing microstrip to stripline or microstrip
to waveguide transitions.
All analyses techniques are employed to calculate the effect of variations in the circuit
parameters on the overall performance. The results are useful for two purposes: optimisation
of the circuit and statistical analysis.

4.7.4 Optimisation
Optimisation is the process of iterative modifications of a set of circuit parameters in order
to achieve a specified performance, i.e. to meet a set of given specifications. The idea of
optimisation is very simple and arises directly from the trial and error tuning procedures of
the classical approach. Basically you alter some values of the circuit parameters and evaluate
the overall performance if it is better you keep the values, otherwise you try a different set.
The advantage of using a computer-aided approach is that the optimisation is performed

Figure 4.72 Typical partitioning of line transformer for planar electromagnetic analysis
Amplifier Design 299

in a systematic manner, thus covering a much larger set of input values than is possible
with any other approach. Although one can envisage a scenario where the design is totally
performed by trial and error with a very powerful computer this is far from the truth even
with today’s most advanced machines. In reality, the huge number of input parameters make
the starting condition and the choice of variables to optimise critical both in terms of actual
performance and design time.
Two critical issues arise when the optimisation of a circuit is attempted:

1. The search method. This identifies the systematic manner in which the optimisable pa-
rameters are altered.
2. The error function formulation. To determine whether the ‘optimum’ performance is
achieved, not only a set of goals is required, but also a formula with which the distance
from the objective is continuously assessed.

When an optimiser executes the chosen search method a new set of parameter values is
calculated. The new values are used to recalculate the error function. The smaller the
obtained value, the more closely the goals are met.
The following paragraphs contain a discussion of the most common search methods and
error function formulas encountered in most RF CAD tools.

4.7.4.1 Optimisation search methods

4.7.4.1.1 Random search


The random optimisers arrive at new parameter values by using a random-number generator
to select a number within the allowed range for each parameter. It is basically a trial and
error process. Starting from an initial set of values with a known error function, a new set of
parameters is obtained using the random numer generator. The error function for the new set
is calculated and compared to the previous. If the result is better, then the new set is used as
the initial point for the iteration, otherwise the new values are discarded and the process is
repeated with the same initial point.

4.7.4.1.2 Gradient search


The gradient search involves the determination of the gradient of the network’s error function.
The advantage of this optimiser is that it progresses much faster to a point where the error
function is minimised, although it is possible that a local minimum is reached rather than the
optimum performance.
A cycle of the gradient optimiser starts with the calculation of the gradient of the error
function for the initial parameter set. This points to a direction that reduces the value of the
error function. Subsequently the initial set is moved in the direction that was determined,
until the error function reaches a minimum. At that point the parameter set becomes the
initial point and the gradient is recalculated.
A single iteration for the gradient search includes several error function evaluations,
thus increasing the required processing time compared to the random. On the other hand,
the gradient optimiser results in a more stable design (small changes of the actual design
parameters do not significantly alter the overall performance).
300 Microwave Devices, Circuits and Subsystems for Communications Engineering

4.7.4.1.3 Quasi-Newton search


The quasi-Newton optimisers employ second-order derivatives of the error function as well
as the gradient in order to determine the direction of the search. The calculated second-order
derivatives supplement the gradient pinpoint more accurately in the direction of the search.
The optimisation terminates when the gradient is zero (a minimum is reached) or when
the estimate variable change becomes to small (less than a predefined value, e.g. 14.68).
Compared to both the gradient and random searches, each iteration of the quasi-Newton
approach requires several more operations.

4.7.4.1.4 Direct search


This is a form of the random optimiser employed when some parameters are allowed to take
only discrete values, for example the resistance of a resistor. This type of search is more of
a trial of permissible combinations of values. So when a set of values is found that results
in a reduction of the error function, it is stored until a better one is determined. Obviously
the optimum performance is determined when all possible combinations are evaluated.
The processing time, however, for such a venture is prohibitive unless a very simple circuit
is investigated.

4.7.4.1.5 Genetic algorithm search


This is a form of direct search employed when the number of discrete parameters is so large
that a direct search of all combinations is impossible. In a sense it is a development of the
previous strategy but instead of blind trial of all combinations, a more intelligent guess is
attempted, by combining some subsets of values rated best in the previous trials.

4.7.4.2 Error function formulation

4.7.4.2.1 Least squares


The least squares error function is calculated by evaluating the error of each specified goal
at the operation range (usually frequency or power) for every point individually and then
squaring the magnitudes of those errors. The error function is then the average of the indi-
vidual errors over the whole range.
A mathematical expression for the error function is shown next:

Summation over all


frequency subgroups
Summation over all Summation over all
optimisation groups frequencies in subgroup
that contains f
Summation of all
responses in subgroup

  ∑ ( ∑ w ⋅ | Rj ( f ) − g j | 2 )  
U = ∑ opt_groups ∑ sub_groups  f i i  (4.82)
  Nf 
Amplifier Design 301

where:
U: is the least square error function
Rj: is the evaluated response at frequency f
gj: is the goal to be achieved for the response Rj
wi: is the weighting factor associated with the response Rj ( f )
Nf : is the number of frequencies of the sub-group to which f belongs.
It is important to point out the significance of the weighting factor. It is evident that any
change in the weightings applied in the various responses would completely alter the value of
the error function. This is significant since it allows the designer to increase the significance
of one response rather than the other. Furthermore the frequency can be substituted with other
swept variables like power or an optimisation for more than one swept valuables can be
attempted. Nevertheless care should be given not to increase excessively the complexity of
the error function as this would result in significant increase of the required processing time.
4.7.4.2.2 Minimax
Minimax optimisation calculates the difference between the desired response over the entire
measurement parameter range of the optimisation. It is the task of the optimiser then to
minimise the difference for the point that has the maximum deviation between the evaluated
and the desired response. In effect, minimax is the minimisation of the maximum of a set of
functions denoted as errors.
The error function is now mathematically represented by the Equation (4.83):
U = max i (Ri − gi) (4.83)
A minimax optimiser always tries to minimise the worst case, giving a solution that has the
goal specifications met in an equal ripple manner. This implies not only that the specifications
should be met, but the error should be smoothed out for all the evaluated responses. In that
sense the minimax solution would be the one that best fits the goals.
4.7.4.2.3 Least Pth
The least Pth error function formulation is similar in make-up to the least squares method
discussed earlier. The difference arises in that the magnitudes of the individual errors are no
longer squared but raised to Pth power, with p = 2,4,8 or 16. This effectively emphasises the
errors that have large values.
As p increases, the least Pth error function approaches the minimax function. The least
Pth formulation is an approximation of the minimax solution. Minimax error function con-
tains infinite gradient changes (discontinuities) when the error contributions due to different
goals intersect in the parameter space. By approximating the minimax with a least Pth, these
discontinuities are smoothed out.
The mathematical formula that describes the least Pth function is given by:
 
1/ p

  ∑ ( Ri − gi ) 
p
if maxi ( Ri − gi ) > 0
 i 

U = 0 if maxi ( Ri − gi ) = 0 (4.84)
 −1/ p
 
  ∑ − ( Ri − gi )  if maxi ( Ri − gi ) < 0
−p
  i 
302 Microwave Devices, Circuits and Subsystems for Communications Engineering

The least Pth method permits the error function to become negative. That implies that even
if a solution is better than the goal (negative error), the solution will be improved further so
that the error would be smooth for all the responses.
Note: It should be evident that the optimisation is not a trivial job, it requires long
computation times especially when the design is complex. Some basic rules that simplify the
task are presented next:

• Choose the optimisable variables sensibly. The complexity increases geometrically as the
number of parameters to examine increases.
• Reduce the parameter space by limiting the input variables. The optimiser is a mathemat-
ical tool and it cannot appreciate the physical significance of the variables.
• Select optimisation goals carefully. Sometimes a non-critical response has to be degraded
to allow an improvement in a more significant area.
• Be aware of circuits with marginal stability.
• Use more than one optimisation technique. For example, gradient optimisation can be
combined with random to ensure that the achieved minimum is a global one.

4.7.5 Further Features of RF CAD Tools


4.7.5.1 Schematic capture of circuits
Schematic capture is one of the major advances in CAD brought about by the new fast graphic
computers. The first CAD tools required a network list (like computer language listing) of
all the elements comprising the circuit where the interconnections where defined by a series
of nodes. The difficulty of introducing circuits with a few hundred nodes is self-evident. A
typical view of a schematic editor is shown in Figure 4.73.
The problem is overcome with the schematic capture. Here the circuit is entered graphically
using icons that represent circuit elements available in the CAD libraries. The additional
benefit is that the designer can introduce the parameters required in a pop-up window
invoked when the new element is positioned. The schematic capture results in a design that
is easy to visualise since it retains the conventional circuit diagram format.
To avoid cluttering the display the concept of hierarchical design was introduced. In this
the design is divided in sub-networks (in a manner similar to block diagrams, each of which
is introduced independently and associated with a unique name and schematic symbol. The
sub-networks can be interconnected together with other elements to create a more complicated
design. The concept is similar to bench working where several instruments and test devices,
each comprising of several electronic circuits are interconnected to create a more complex
circuit.

4.7.5.2 Layout-based design


To further simplify the fabrication process modern RF CAD tools allow the designer to con-
vert schematic diagrams into layout representations of the circuit. This capability allows for
quick generation of the artwork required for the manufacturing of RF designs. Figure 4.74
contains the layout representation of the schematic shown in Figure 4.73. In essence it is a
form of schematic representations since it can allow almost the same features. For example,
Amplifier Design 303

Figure 4.73 Schematic representation of a power amplifier

a circuit can be designed, simulated, and optimised completely in the layout form. This is
achieved by associating each component available in the library with three representations:

• a schematic representation;
• an artwork representation;
• an electrical model.

In that sense the correspondence is direct, when the simulator encounters one, it can sub-
stitute it with the other. This allows direct conversion from the electrical design (schematic
representation) to an executable program (for analysis purposes) or an artwork (layout
representation). The direct transfer of data from an easy-to-read schematic representation
to an artwork not only reduces the actual design time, but minimises the probability of an
error creeping in at later stage.

4.7.5.3 Statistical design of RF circuits


Statistical design is the process of:

• accounting for the random variations of the parameters of a design;


• measuring the effect on the circuit performance of such variations;
• modifying the design to minimise these effects.
304 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 4.74 Layout representation of a power amplifier

In a practical application where RF circuits are mass-produced, it is important to identify


the effect of the statistical variation of the nominal values of the individual components.
In essence it is necessary to identify what are the acceptable deviations of the nominal value
(tolerance) so that the resulting performance will lie within the specified range. The process
of varying a set of input parameters using specified probability distributions to determine
what percentage of resulting designs fall within the specifications is called yield analysis.
The ratio of the number of circuits that pass the design specification to those produced is
called the yield of the design.
To determine the yield of a circuit, one needs to identify which are the limits within which
the performance of the actual design is considered acceptable. Figure 4.75 depicts a typical
frequency response diagram which depicts the limits of this region (shaded area). To simplify
the process let’s assume that only two input parameters P1 and P2 can affect the outcome
(the tolerances of these parameters, ∆P1∆P2 , are displayed in Figure 4.76). By analysing the
circuit for all the combinations of P1 and P2 we can determine a region that contains all the
sets of P1 and P2 that result in acceptable performance, called the element constraint region
(depicted in Figure 4.76). To obtain high yield the input parameters have to remain within
the limits of this area. The yield of the circuit is the superimposed area.
To maximise the yield it is necessary to maximise the overlap area in the tolerance
rectangle and the element constraint region. This can be done by shifting the tolerance
square in the parameter space until the two are centred (Figure 4.77). This process is called
Amplifier Design 305

Responce

Figure 4.75 Range of acceptable performance for the designed circuit

Element constrained
region

∆P2
Parameter P2

∆P1

Parameter P1

Figure 4.76 Determination of the yield for a given tolerance window

∆P2
Parameter P2

∆P1
Parameter P1

Figure 4.77 Design centring: maximisation of the circuit yield by centring the tolerance window
and the element constrained region
306 Microwave Devices, Circuits and Subsystems for Communications Engineering

yield optimisation or design centering. If fabrication yield is needed, it can be seen that more
stringent limits must be set on the tolerances of the input parameters. It is the designer’s task
to determine whether the reduction of the cost by the minimisation of the failed circuits
justifies the additional cost of obtaining components with higher tolerances.

Appendix I: Second Type of Circle Mapping


Suppose we have an expression of the form:

A x 2 + Ay 2 + Bx + Cy + D
X=
Ex 2 + Ey 2 + Fx + Gy + H

where A to F are a set of real constants. If we write a complex number z as z = x + jy, then
this function generates a real number X associated with any point in the complex z plane. It
is easy to show that the loci of constant X in the z plane are circles. The distinguishing
feature of the expression which guarantees the circle property is that the coefficients of x 2
and y 2 in the numerator are the same (both A) and likewise in the denominator (both E).
Without this condition, the loci of constant X would become ellipses.

Appendix II: Masons’s Rule and Signal Flow Graphs


Signal flow graph techniques provide a graphical representation of the relationships between
circuit parameters to which systematic manipulations can be applied to determine the per-
formance of the network. The general solution to a signal flow graph can be obtained using
Mason’s rule.
Example networks are shown in Figure AII.1, where the networks consist of directed
branches between nodes. Each branch has a branch transmittance that describes the relation-
ships between the signals at the node signals seen at each end of the branch. A node signal
is equal to the algebraic sum of all signals entering that node, and that signal is applied to
each of its outgoing branches.

(a) (b)
a1 p a1 S21
ΓG ΓL S11 S22 ΓL
q b1 S12

(c)
a1 L21 M21 b2
L11 L22 M11 M22
b1 L12 M12 a2

(d)
a1 L21 M21 N21 b2
L11 L22 M11 M22 N11 N22
b1 L12 M12 N12 a2

Figure AII.1 Example cascade circuits for analysis by Mason’s Rule


Amplifier Design 307

AII.1 Mason’s Non-Touching Loop Rule


Masons rule is a general method of determining the transmittance of an overall circuit from
its flow graph. In these diagrams we have sources where the node has only outgoing branches,
such as a1 and a2, and sinks where the node has only incoming branches, such as b1 and b2.
Paths are formed by a continuous succession of branches that follow the direction of the
arrows, in which any node is encountered only once.
A closed path that loops back to its starting node is a loop, and these are further charac-
terised as:

• First order loop: a closed path looping back to a node without crossing the same node twice.
• Second order loop: the product of any two non-touching first order loops.
• Third order loop: the product of any three non-touching first order loops.
• Fourth order loop, etc . . .

The path transmittance, P, is the product of the branch transmittances which make up the
path, while the loop transmittance, L, is the product of the branch transmittances which
make up the loop. The graph transmittance, K, is the ratio of the signal appearing at the sink
node to the signal applied from the source node (i.e. the overall desired circuit transmittance).
Using the above definitions, Mason’s rule defines K as:

1 n
K= ∑ Pi ∆i
∆ i =1
(A1)

where :

n = the number of forward paths from source to sink


∆ = 1 − (Sum of all first order loops) + (sum of all second order loops) − (sum of all third
order loops) + (sum of all fourth order loops) − ( . . .
Pi = transmittance of the ith forward path
∆ i = the value of ∆ for that portion of the graph not touching the ith forward path

AII.1.1 One Port


In this case, Figure AII.1(a), there is a single loop ΓGΓL and a unit transmittance path to the
node signal p, giving

1
p = a1 (A2)
1 − ΓG ΓL

AII.1.2 Two-Port – Single Element


Considering this circuit, Figure AII.1(b), we can identify the paths and loops:

Paths: a1 to b1 P1 = S11, P2 = S21ΓL S12


Loops: First Order L1 = ΓL S22
Second Order None
308 Microwave Devices, Circuits and Subsystems for Communications Engineering

In this case ∆ = 1 − S22ΓL. The loop L1 does not touch path P1 so ∆1 = 1 − S22 ΓL, but L1
does touch path P2 and ∆2 = 1. The overall transmittance from a1 to b1, or input reflection
coefficient Γi, is then:
P1∆1 + P2 ∆ 2 S (1 − S22 ΓL ) + S21S12 ΓL S S Γ
K = Γi = = 11 = S11 + 21 12 L (A3)
∆ (1 − S22 ΓL ) 1 − S22 ΓL

AII.1.3 Two-Port – two Element Cascade


Considering the circuit, Figure AII.1(c), we identify the paths and loops:

Paths: a1 to b1 PA1 = L11, PA2 = L21 M11 L12


a1 to b2 PB1 = L21 M21
a2 to b1 PC1 = L12 M12
a2 to b2 PD1 = M22, PD2 = M12 L22 M21
Loops: First Order L1 = L22 M11
Second Order None

For all paths we have ∆ = 1 − L22 M11.


The loop LA1 does not touch path PA1 so ∆ = 1 − L22M11, but LA1 does touch path PA2 and
∆2 = 1. The overall transmittance from a1 to b1, or as we are considering a2 = 0 in these
equations the overall S11 of the circuit, is then:
PA1∆ A1 + PA2 ∆ A2 L (1 − L22 M11 ) + L21 M11 L12 L L M
S11′ = = 11 = L11 + 21 12 11 (A4)
∆ (1 − L22 M11 ) 1 − L22 M11
The forward transmission through the circuit from a1 to b2 has the path, PB1, which touches
L1 so ∆B1 = 1. The overall S21 of the circuit is:
PB1∆ B1 L21 M21
′ =
S 21 = (A5)
∆ (1 − L22 M11 )
Likewise, for the reverse transmission from a2 to b1 we have the path PC1, and
PC1∆ C1 L12 M12
′ =
S12 = (A6)
∆ (1 − L22 M11 )
The last path is similar to the first and gives the S22 of the circuit. The loop LD2 does not
touch path PD1 so ∆D1 = 1 − L22M11, but LD1 does touch path PD2 and ∆D2 = 1. Hence:
PD1∆ D1 + PD2 ∆ D2 L (1 − L22 M11 ) + L12 M22 L21 L L M
′ =
S 22 = 22 = L22 + 12 21 22 (A7)
∆ (1 − L22 M11 ) 1 − L22 M11

AII.1.4 Two-Port – three element cascade


From the circuit in Figure AII.1(d) we identify the loops and paths:

Paths: a1 to b1 PA1 = L11, PA2 = L21M11 L12, PA3 = L21M21N11M12 L12


a1 to b2 PB1 = L21 M21 N21
a2 to b1 PC1 = L12 M12 N12
a2 to b2 PD1 = N22, PD2 = N12 M22 N21, PD3 = N12 M12 L22 M21 N21
Amplifier Design 309

Loops: First Order L1 = L22 M11, L2 = M22 N11, L3 = L22 M21 N11 M12
Second Order L21 = L22 M11 M22 N11
Third Order None

from which we can determine the overall S-parameters of the cascade circuit.
For all paths we have: ∆ = 1 − L1 − L2 − L3 + L21.
For the paths from a1 to b1 we can identify:

PA1 is not touched by any loop, so ∆A1 = ∆.


PA2 is touched by L1 and L3, so ∆A2 = 1 − L2.
PA3 is not touched by all loop, so ∆A3 = 1.

Hence the reflection parameter for the cascade is:

PA1∆ A1 + PA2 ∆ A2 + PA3∆ A3 P ∆ + PA2 (1 − L2 ) + PA3


S11′′ = = A1
∆ ∆
PA2 (1 − L2 ) + PA3 PA2 (1 − L 2 ) + PA3
= PA1 + = PA1 +
∆ 1 − L1 − L 2 − L 3 + L 21
L21M11L12 (1 − M22 N11 ) + L21M21 N11 M12 L12
= L11 +
1 − L22 M11 − M22 N11 − L22 M21 N11 M12 + L22 M11 M22 N11
L21L12 ( M11(1 − M22 N11 ) + M21 N11 M12 )
= L11 +
1 − M22 N11 − L22 ( M11 + M21 N11 M12 − M11 M22 N11 )
L21L12 ( M11(1 − M22 N11 ) + M21 N11 M12 )
= L11 +
1 − M22 N11 − L22 ( M11(1 − M22 N11 ) + M21 N11 M12 )
 M N M 
L21L12  M11 + 21 11 12 
 1 − M22 N11 
= L11 + (A8)
 M N M 
1 − L22  M11 + 21 11 12 
 1 − M22 N11 
In the case of the second set of paths from a1 to b2 the path PB1 is touched by all loops, so
∆B1 = 1. The transmission parameter for the cascade is therefore:

PB1∆ B1 L21M21 N21


′′ =
S21 =
∆ 1 − L22 M11 − M22 N11 − L22 M21 N11 M12 + L22 M11 M22 N11
L21M21 N21
=
1 − M22 N11 − L22 ( M11 + M21 N11 M12 − M11 M22 N11 )
L21M21 N21
=
1 − M22 N11 − L22 ( M11(1 − M22 N11 ) + M21 N11 M12 )
L21M21 N21
1 − M22 N11
= (A9)
 M N M 
1 − L22  M11 + 21 11 12 
 1 − M22 N11 
310 Microwave Devices, Circuits and Subsystems for Communications Engineering

References
[1] H.W. Bode, Network Analysis and Feedback Amplifier Design, Van Nostrand, New York, 1945.
[2] R.M. Fano, ‘Theoretical limitations on the broad-band matching of arbitrary impedances’, Journal of the Franklin
Institute, Vol. 249, pp. 57–83, January 1950, and pp. 139–154, February 1950.
[3] I.D. Robertson (ed.), MMIC Design, IEE, London, 1995.
[4] CDS Series IV, Simulating and Testing, Hewlett Packard, New Jersey, 1995.
[5] CDS Series IV, Momentum User’s Guide, Hewlett Packard, New Jersey, 1995.
[6] G.D. Vendelin, A.M. Pavio, U.L. Rohde, Microwave Circuit Design Using Linear and Nonlinear Techniques,
Wiley & Sons, New York, 1992.
[7] K.C. Gupta, R. Garg, R. Chadha, Computer Aided Design of Microwave Circuits, Artech House, 1991.
[8] T.C. Edwards, Foundations for Microstrip Circuit Design, Wiley & Sons, New York, 1992.
[9] S.R. Pennock, P.R. Shepherd, Microwave Engineering for Wireless Applications, Macmillan, Basingstoke, 1998.
[10] G. Kaplan, ‘Special guide to software systems, packages and applications’, IEEE Spectrum, November 1990,
pp. 47–101.
[11] C. Bowick, RF Circuit Design, H.W. Sams & Co., London, 1982.
[12] D.M. Pozar, Microwave Engineering, Reading, MA, Addison-Wesley, 1990.
[13] S.Y. Liao, Microwave Circuit Analysis and Amplifier Design, Prentice Hall Inc., Englewood Cliffs, NJ, 1987.
[14] F.A. Benson, T.M. Benson, Fields Waves and Transmission Lines, Chapman & Hall, London, 1991.
[15] G.D. Vendelin, A.M. Pavio, U.L. Rohde, Microwave Circuit Design Using Linear and Nonlinear Techniques,
Wiley & Sons, New York, 1992.
Mixers: Theory and Design 311

5
Mixers: Theory and Design
L. de la Fuente and A. Tazon

5.1 Introduction
Crystal detectors and mixers are the key circuits in receiver systems. At the start of the
twentieth century detectors were very unreliable, they were built using a semiconductor crystal
and a thin wire contact that had to be periodically adjusted to guarantee good behaviour. A
significant advance in receiver sensitivity was the development of triodes since they made
it possible to amplify before and after the detector. However, the real advance was obtained
by the invention of the superregenerative receiver by Major Edwin Armstrong. Armstrong
was also the first to use the vacuum tube as a frequency converter (mixer) to change the
frequency from the received signal (RF) to an intermediate frequency (IF), which could be
amplified, selected with low noise and detected. Armstrong is considered to be the inventor
of the mixer. This kind of receiver (superheterodyne receiver) has been, up to now, the
greatest advance in receiver architecture and it is used in practically all modern receivers.
During the Second World War the development of the microwave mixers took place due
to radar development. At the beginning of the 1940s, single-diode mixers had very poor
noise figures but by the 1950s, it was possible to obtain noise figures around 8 dB. Nowadays,
mixers have this behaviour at frequencies greater than 100 GHz. Mixer theory has been
developed and it is possible to reach noise figures of less than 4 dB up to 50 GHz.
Figure 5.1 shows a simplified scheme of a double-mix superheterodyne receiver. The
signal received (RF) by the antenna is filtered, amplified by a low noise amplifier and mixed
with the frequency of the first local oscillator to obtain the first intermediate frequency (1st
IF). This IF is also amplified and filtered and mixed with the frequency of a synthesised
oscillator (second local oscillator) to obtain the second intermediate frequency (2nd IF). This
2nd IF (base band) is filtered, amplified and detected to obtain the information.

5.2 General Properties


As we saw in the previous section, mixers are basic circuits in the emitter and receiver
systems. A mixer is basically a multiplier, and if we suppose two sinusoidal signals at the
multiplier input, we obtain the product of these signals at the output, as shown in Figure 5.2.

Microwave Devices, Circuits and Subsystems for Communications Engineering Edited by I. A. Glover, S. R. Pennock
and P. R. Shepherd
© 2005 John Wiley & Sons, Ltd.
312 Microwave Devices, Circuits and Subsystems for Communications Engineering

MIX1 MIX2 LPF

RF
BPF LNA LPF AMP LPF AMP
2nd IF
LO1 1st IF
COUPLER
LO2 DET

P.L.L.

PLLcontrol

Figure 5.1 Double-mix synthesized receiver

Ideal Mult.
LPF
A(t)·cos ωRFt IF IF
RF A(t)·cos (ωRFt - ωLOt)/2

LO cos ωLOt

A(t)·cos ωRFt·cos ωLOt = 1/2·A(t)·[cos (ωRF - ωLO)t + cos (ωRF + ωLO)t]

Figure 5.2 Signal components in an ideal mixer

Figure 5.2 shows an ideal mixer. The carrier is the RF signal and it is mixed with the LO
signal and the information is transmitted by A(t). The response of the multiplier is two
sinusoidal IF components (mixing products). Normally, in communication systems, only one
component is desired while the other is rejected by using appropriate filters.
However, electronic devices such as ideal multipliers do not exist, which provokes prob-
lems such as generation of superior harmonics and spurious intermodulation signals due to
the non-linearity of the mixer. These signals, called spurious responses, must be eliminated
later by filtering.
The mixers, even if they are ideal mixers, have a second frequency (2fLO − fRF) that can
give an IF response. This kind of spurious response is called image frequency. For instance, if
we suppose a typical VSAT reception frequency fRF = 13.5 GHz and a frequency of the local
oscillator fLO = 12.3 GHz, we obtain an intermediate frequency fIF = 13.5 − 12.3 = 1.2 GHz
(typical value of the first IF). However, the non-linearity of the mixer gives the third band
intermodulation product 2fLO − fRF = 24.6 − 13.5 = 11.1 GHz. The mixer can give, fLO − 11.1
= 1 GHz, a spurious signal whose frequency is the same as IF. It is possible to develop
hybrid circuits or combinations of mixers capable of rejecting the image response.
Mixers: Theory and Design 313

5.3 Devices for Mixers


In theory, any non-linear electronic device can be used as a mixer, however, there are only
a few devices that have the practical requirements of mixers in communication systems. A
device used as a mixer must satisfy the following characteristics:

• strong non-linearity;
• reliability and low dispersion of the devices;
• low noise;
• low distortion;
• good frequency response.

5.3.1 The Schottky-Barrier Diode


The Schottky-Barrier diode is most commonly used in mixer circuits and it is basically a
metal-semiconductor junction. As well as its good behaviour as a mixer, the success of the
Schottky-Barrier lies in that it can reach very high frequency (up to 1000 GHz) and it is
relatively cheap. In the past, PN-junction diodes, mainly point-contact diodes, were used
in mixers. Point-contact diodes are formed by pressing a contact wire onto a piece of semi-
conductor. This produces a primitive Schottky-Barrier or metal-to-semiconductor junction.
However, the building process of modern Schottky-Barrier diodes (photolithography over
an epitaxial substrate) gives more reliability than the point-contact diodes. The good char-
acteristics of Gallium Arsenide (GaAs) Schottky-Barrier diodes have led to an important
development of millimetre-wave mixers as GaAs has greater electron mobility and saturation
velocity than silicon semiconductors. Figure 5.3 shows a Schottky-Barrier diode (a) and its
equivalent circuit (b).

5.3.1.1 Non-linear equivalent circuit


The typical non-linear equation of the current source i(v) is:

[ ]
q⋅v
i(v) = Iss ⋅ e n ⋅ K ⋅ T − 1 (5.1)

where Iss is the saturation current, q is the electron charge, K is the Boltzmann’s constant,
T is the absolute temperature and n is a parameter called ideality factor whose value is
between 1.0 and 1.25.

i i

ve i(v) v cj(v)

Rs
(a) (b)

Figure 5.3 Schottky-Barrier diode equivalent circuit


314 Microwave Devices, Circuits and Subsystems for Communications Engineering

i(v) cj(v)

cj0

v v
(a) (b)

Figure 5.4 (a) Schottky diode current source characteristic


(b) Schottky diode junction capacitance characteristic

The non-linear junction capacitance can be expressed by the equation:

Cj 0
Cj = γ
(5.2)
 v
1 − 
 φbi 

where, Cj0 is the junction capacitance at zero bias voltage, φbi is the built-in voltage and
its value typically varies between 0.7 and 0.9 volts, and γ is a parameter that varies between
0.5 and 0.33.
The series resistor of Figure 5.3 Rs is the loss resistance and its value is constant, around
a few ohms. Figure 5.4(a) shows the typical curve of the current source i(v) and Figure 5.4(b)
shows the characteristic of the junction capacitance Cj (v).

5.3.1.2 Linear equivalent circuit at an operating point


It is possible to deduce the small signal equivalent circuit of the current source (Figure 5.4a)
if we suppose that v is a very small sinusoidal voltage around a DC voltage V0 (operating
point) as we can see in Figure 5.5, where the input signal around the operating point
(V0, I0), is:

v = V0 + V1 · cos ω 0 · t (5.3)

i(v)

I0
i
V0 v
v

Figure 5.5 Linearization of the Schottky diode i(v) characteristic


Mixers: Theory and Design 315

The approximated response of the current source i(v) can be obtained by the first term of the
Taylor series:

i(v) = i(V0 + V1 · cos ω 0 · t) =

∂i
= i(V0) + ·v +...=
∂ v V0
= I0 + g(V0) · v = I0 + i

where:

i = g(V0) · v (5.4)

Equation (5.4) is a linear law (Ohm’s Law) and the transconductance, taking into account
Equation (5.1), can be written as:

∂i q q⋅v
q
g(v) = = ⋅ Iss ⋅ e n ⋅ K ⋅ T = ⋅ [i(v) + Iss] (5.5)
∂v n ⋅ K ⋅ T n⋅K ⋅T

The linear representation of the current source i(v) and its linear spectral response can be
seen in Figure 5.6.
In the case of the capacitance the problem is different because the non-linear characteristic
is Q(v) and the current response is given by Equation (5.6).

dQ ∂Q dv dv
i(t ) = = ⋅ = C j (v) ⋅ (5.6)
dt ∂v dt dt

Where Cj (v) is given by Equation (5.2).


When sinusoidal excitation of Equation (5.3) is very small, the approximate response of
Q(v) can be obtained using the first term of the Taylor series (Equation 5.7):

Q(v) = Q[V0 + V1 · cos(ω 0 t)] = Q(V0) + Cj (V0) · V1 · cos(ω 0 t) (5.7)

and the approximate linear response is given by Equation (5.8):

dQ d π
i(t ) = = [Q(V0 ) + C j (V0 ) ⋅ V1 ⋅ cos (ω 0t )] = C j (V0 ) ⋅ ω 0 ⋅ V1 ⋅ cos (ω 0t + ) (5.8)
dt dt 2

i
Amp.
v =V0 + v = V0 +V1*cos(ω 0t)
g(V0).
i(v) v
( Around
(V , I ) )
0 0
g v

ω0
Non-Linear (a) Linear (b)

Figure 5.6 (a) Source current equivalent circuit (b) Spectral representation
316 Microwave Devices, Circuits and Subsystems for Communications Engineering

Cj(v)
Q(v)
Cj0
C(V0)
Cj (V0)

V0 v V0 v
(a) (b)

Figure 5.7 (a) Non-linear charge characteristic (b) Linearization of the Schottky
capacitance at V0 DC voltage

i
Amp. Cj(V0).ω 0.V1
v =V0 + v = V0 +V1*cos(ω 0t)

Cj(v) v
( Around
(V , I ) )
0 0
v

Cj(V0) ω0
Non Linear (a) Linear (b)

Figure 5.8 (a) Schottky capacitance equivalent circuit (b) Spectral representation

i=I0+i

g v Cj(V0)

v=V0 + v
(Around
(V , I ) )
0 0 Rs

Figure 5.9 Linear equivalent circuit of Schottky diode

As we can see in Equation (5.8), there is no DC term and the amplitude of the ω 0 response
is: Cj (V0) · ω 0 · V1. Figure 5.7 shows the linearisation of any non-linear charge characteristic
and the Schottky junction capacitance. The linearisation of the equivalent circuit of the
Schottky capacitance and the spectral representation can be seen in Figure 5.8 where Cj (V0)
is given by Equation (5.9). Therefore, the total linear circuit of a Schottky diode around V0
operation point is given by Figure 5.9.

Cj 0
C j (V0 ) = γ
(5.9)
 V0 
1 − 
 φ
Mixers: Theory and Design 317

i(µ A) Approx.

1000 Real

100 ∆Ve = 0.017 V


i2
10
i1
1
∆Ve = 0.062 V
0.1
V1 V2
0.4 0.5 0.6 0.7 0.8 Ve (Volts)

Figure 5.10 Experimental characteristic of a Schottky diode obtained by a curve tracer

5.3.1.3 Experimental characterisation of Schottky diodes


Taking into account the non-linear equivalent circuit of Figure 5.3, we can write the DC
diode equation as:
ve = v + i(v) · Rs (5.10)
and a semi-logarithmic representation of the diode characteristic obtained by a curve tracer
is given in Figure 5.10. In the zone where i is very small but greater than Iss and taking into
account Equation (5.1) we can write:
q⋅v
log i ≈ log Iss + ⋅ log (e) (5.11)
n⋅K ⋅T
Considering Equation (5.11) and taking into account the points i1 and i2 = 10 · i1 of
Figure 5.10, we can write:
q ⋅ ∆ve
log (i2 ) − log (i1 ) = 1 = ⋅ log (e) (5.12)
n⋅K ⋅T
and therefore:
q ⋅ ∆ve ⋅ log (e) ∆ve
n= ≅ (5.13)
K ⋅T 0.05783
K ⋅T
because ≈ 0.025 Volts at ambient temperature. In this case, the ideality is n = 1.07.
q
When i(v) · Rs is very small, Equation (5.10) represents the straight part of the diode
characteristic of Figure 5.10 and, at any point of this part, we can obtain:
q⋅v
− n⋅ K ⋅T
Iss = I (v) ⋅ e (5.14)
−13
In this case, Iss = 1.25 10 A.
Taking into account the deviation between the real and approximate diode characteristic
of Figure 5.10, we can calculate Rs as:
ve = v + Rs · i(v) (5.15)
318 Microwave Devices, Circuits and Subsystems for Communications Engineering

1/Cj2
300
1/C 2j0 =120 => Cj 0 = 0.091pF
200

100 Φ = 0.8 V

-1.0 -0.5 0.5 1.0 1.5 Ve (Volts)

Figure 5.11 Schottky diode capacitance characteristic obtained by


an impedance meter at 1 MHz

and

ve − v
Rs = (5.16)
i( v )

Where ∆Ve = ve − v = 0.017 V (Figure 5.10) and therefore, Rs = 17 Ω.


On the other hand, taking into account the non-linear capacitance of the diode equivalent
circuit of Figure 5.3, it is possible to obtain, by an impedance bridge meter, the characteristic
of Figure 5.11.
From Equation (5.2) and Figure 5.11 we can obtain the Schottky junction capacitance as:

1 1  v 
= 2 ⋅ 1 − (5.17)
2
Cj C j0  φbi 

Where φbi = 0.8 Volts and Cj0 = 0.091 pF, in Figure 5.11.
An important merit figure is the cut-off frequency. If we suppose, in Figure 5.3, a voltage
around zero, we can write the diode impedance as:

j
Zd = RS − (5.18)
ω ⋅ Cj 0
the phase relationship is:

V V
I= ⇒ Imax = (5.19)
j RS
RS −
ω ⋅ Cj 0

As the cut-off frequency is defined at the 3 dB point we can write:

I V  1  Imax 1
= =  RS = = ⇒ fC = (5.20)
2 j  ω C ⋅ C j0  | 1 − j | 2 ⋅ π ⋅ RS ⋅ C j 0
RS −
ω C ⋅ Cj 0
Mixers: Theory and Design 319

Another important merit figure is the quality factor. Taking into account that the circuit of
Figure 5.3 around v = 0 volts is a RS Cj 0 series circuit, we can deduce the quality factor as:
1 f
Q= = C (5.21)
2 ⋅ π ⋅ f0 ⋅ RS ⋅ C j 0 f0

5.3.2 Bipolar Transistors


Bipolar transistors (BJTs) have been used as mixers up to a few GHz but the development
of heterojunction bipolar transistors (HBTs) at higher frequencies has increased the interest
of these devices in the design of RF and microwave mixers. The bipolar transistors are
also known as homojunction bipolar transistors but as the initials coincide with the HBT
devices, they are called BJTs. BJTs are made in Silicon (Si) technology because it is diffi-
cult to control the p-type particles in Gallium Arsenide (GaAs) technology. Furthermore,
this technology would need a weakly doped p-type to obtain high gain and this means a very
high base resistance. All these problems lead to GaAs BJTs having worse behaviour than
Si devices.
The great importance of the HBT devices in recent years is due to the high transconductance
and output resistance values reached by these devices, along with the high power capability
and the high breakdown voltages. Moreover, HBTs admit very low base resistance values
with wide emitter terminals and these properties allow these transistors to reach very high
operation frequencies. Thus, the HBTs are very useful devices for millimetre wave applications
with high power levels.
In conclusion, we can say that the Si bipolar transistors (BJT) are used up to a few GHz,
they use low cost technology and have high performance. GaAs and SiGe HBTs are used
at high frequency. However, the HBT and BJT equivalent circuits and their I/V and Q/V
characteristics are similar. The small differences between the two devices are:

• High doping densities of HBTs require less base width than BJTs and the high injection
effects practically do not exist.
• Low current region in HBTs is greater than the same region in BJTs, this effect means
that β increases monotonously with the collector current.
• The base current is less.

In most applications, the non-linearity base-emitter diode function is employed for mixing
applications. Taking into account the generic non-linear model of Chapter 2, under forward
bias conditions (Vbe ≥ 0 and Vbc < 0), the device behaviour can be approximated by the
simplified circuit of Figure 5.12. This equivalent circuit along with the Gummel-Poon forward
bias model permits the deduction of the currents through the collector and base terminals.
These currents are given by Equation (5.22a, 5.22b):

  V  
Icc = Isf  exp  be  − 1 (5.22a)
  n f KT  
  V  
Ibe = Ise  exp  be  − 1 (5.22b)
  ne KT  
320 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 5.12 DC simplified equivalent circuit under forward bias condition of a HBT

Figure 5.13 One diode simplified circuit under forward bias conditions

where the current source Icc (eq. 5.22a) is the collector current while the base current is
represented by two diodes whose current sources are: Ibe (eq. 5.22b) and Icc /βf . βf is the
forward DC gain of the device.
The equivalent circuit of Figure 5.12 can be simplified by the equivalent circuit of
Figure 5.13. In this case, the diode function can be represented by Equation (5.23):

  V  
IbeT = IseT  exp  be  − 1 (5.23)
  nT KT  

where:

IbeT is the total base-emitter current source.


IseT is the equivalent saturation current.
NT is the equivalent ideality factor.

The non-linearity base-emitter equivalent diode function is mainly responsible for the
mixing function.
Mixers: Theory and Design 321

At low RF frequencies, the inductive and capacitive effects can be ignored but the access
resistances of the HBTs and BJTs (Figure 2.58 of Chapter 2), Rb, Rc and Re, play an import-
ant role in the behaviour of these devices because the experimental I/V and the base-emitter
diode function characteristics of the transistor are measured as a function of the external
voltages Vc and Vb but the voltages of the Equations (5.22) and (5.23) are functions of the
internal (intrinsic) voltages.
These access resistances can be obtained from considerations of the geometrical and
material properties or experimental measurements. The relationship between internal and
external voltages is given by equations (2.148) and (2.149) of Chapter 2.
At high frequencies, access inductances and linear and non-linear capacitances (Figure
2.58 of Chapter 2) must be taken into account. Traditional extracting methods from scattering
parameter measurements at different bias points can be employed using the small signal
equivalent circuit of Figure 2.62 (Chapter 2).

5.3.3 Field-Effect Transistors

The Field-Effect Transistors (FET) currently play an important role in the RF and micro-
wave circuit development, including mixer circuits for communication system applications.
Furthermore, the monolithic technology has increased the importance of these devices.
Basically there are three kinds of field effect devices whose differences of behaviour is
the used technology:

1. Metal Oxide Semiconductor FET (MOSFET): This is developed in silicon (Si) technology
and its main characteristic is its power capabilities, mainly for amplification applications.
These transistors find applications in mobile communications where their frequency bands
can reach several GHz. These devices are not especially used in mixer design.
2. MESFET: Silicon (Si) and Gallium Arsenide (GaAs) technologies are normally used
in these type of transistors. Si technology can be used up to several GHz while GaAs
technology can reach up to the millimetre bands. MESFET devices are employed in the
development of different circuits for communications systems including mixing function,
mainly up to Ku band.
3. HEMT (High Electron Mobility Transistor): The most recently developed field effect
transistor is the HEMT. These devices are heterostructures in GaAs technology and the
main difference with respect to the traditional FET is the transconductance compression.
It has very good features of low noise and high gain at high frequencies (Ku, Ka and
higher frequencies). HEMT devices are very usual in mixer applications mainly at Ka
band and above.

If we analyse the three types of field effect transistors, we can deduce that MESFET
and HEMT are the most interesting devices for mixing applications although, at high
frequencies, HBTs (up to Ka band in GaAs technology and up to millimetre frequencies
in SiGe technology) are a very important alternative. Also, dual-gate FETs are widely used
as mixers because these devices have certain advantages with respect to conventional FET
mixers:
322 Microwave Devices, Circuits and Subsystems for Communications Engineering

• local Oscillator (LO) and RF signals can be applied in separated gates with an intrinsic
isolation of 20 dB without filters or balanced structures;
• they have very linear transconductance and present low distortion.

Although there are differences between MESFET and HEMT devices, the equivalent circuit
topology is basically the same for all types of FETs and the differences are apparent in the
values in the equivalent circuit parameters. The most important non-linearity of the FET
device for mixing is the channel current source Ids (Figure 2.52 in Chapter 2). Another
important non-linearity is the gate-to-source capacitance Cds but its main effect is the limita-
tion of useful bandwidth. The rest of the resistive and reactive parasitic elements of the FET
equivalent circuit are of secondary interest in mixing applications.
The behaviour of the current source Ids when we introduce two RF signals is similar to a
near ideal multiplier as we saw in Section 5.2 (Figure 5.2). The Curtice Ids non-linear current
equation is given by:

Ids(Vgi, Vdi) = β (Vgi − VTO)2(1 + λVdi) tanh (αVdi) (5.24)

where two signals (LO and RF) have been applied at the saturation region at any DC
operating point, the RF behaviour of Equation (5.26) can be approximated by:

Ids = β · (vgi − VTO)2 (5.25)

In this case: vgi = vRF + vLO

where:

vRF = A · cos(ωRF · t) RF signal


vLO = B · cos(ωOL · t) LO signal

and the current source of Equation (5.27) is given by:

Ids = β · [A′ cos(ωRF · t) + B′ · cos(ωOL · t) − VTO]2 = a0 + a1 · cos(ωRF · t) +


+ a2 · cos(ωOL · t) + a3 · cos(2 · ωRF · t) + a4 · cos(2 · ωOL · t) + (5.26)
+ a5 · cos(ωRF · t) · cos(ωOL · t)

The last term of Equation (5.26) contains the product function that represents the sum
or difference frequencies (mixing function). The rest of the terms can be eliminated by
filtering.

5.4 Non-Linear Analysis


As we saw in Figure 5.2, a mixer is a multiplier and the non-linear devices used for this
operation were studied in Section 5.3. The common mixing behaviour is basically frequency
conversion in a time-varying parameter of a non-linear device. This parameter appears by
applying a strong waveform to a non-linear device. In this part we will assume a general non-
linear characteristic to calculate the intermodulation products when we apply two sinusoidal
voltages. One of them, the LO, is stronger than the other one, the RF signal.
Mixers: Theory and Design 323

i(v) v(t) =vLO +vS

Figure 5.14 Non-linear characteristic

5.4.1 Intermodulation Products


The characteristic i(v) of the non-linear device in Figure 5.14 can be described by:

i(v) = a0 + a1 · v + a2 · v 2 + . . . = ∑a
k =0
k ⋅ vk (5.27)

The input signal at the circuit of Figure 5.14 is given by v(t) = vLO(t) + vS (t), where the local
oscillator signal (LO) is:

vLO(t) = VP · cos(ωP · t + φP) (5.28)

and the RF signal:

vS (t) = VS · cos(ωS · t + φS) (5.29)

To simplify, we suppose that φP = φS = 0. This does not mean a loss in generality of the
study. In this case, the input signal is given by:

v(t) = vLO(t) + vS (t) = VP · cos(ωP · t) + VS · (ωS · t) (5.30)

If we suppose that Vp Ⰶ VS and developing in Taylor series around vLO the characteristic i(v)
can be written as:

1  di  1  d 2i 
i(v) = i(vLO + vS) = i(vLO) + ⋅ ⋅ vS + ⋅  2  ⋅ vS2 + . . . =
1!  dv  vS = 0 2!  dv  vS = 0
VS2
= i(vLO) + VS · cos(ω S · t) · g(1)(vLO) + · cos2(ω S · t) · g(2)(vLO) + . . . (5.31)
2

where:

di d 2i d ni
g( 0 ) (v) = i(v) g(1) (v) = g ( 2 ) (v) = 2
. . . g ( n ) ( v) = n (5.32)
dv dv dv

so:

VSn
i(v) = ∑
n=0 n!
· cosn(ω S · t) · g(n)(vLO) (5.33)
324 Microwave Devices, Circuits and Subsystems for Communications Engineering

where:

d ni dn  ∞  ∞
k!
n ∑ k ∑ (k − n)! ⋅ a ⋅ v
g( n) (vLO ) = = a vk  = k
k −n
LO (5.34)
dv n vLO dv  k = 0  v =vLO k =n

[vLO = VP · cos(ω P · t)]

Applying the variable change p = k − n in Expression (5.34), we can obtain the conductance
g(n) as:

( p + n)!
g( n) (vLO ) = ∑
p=0 p!
⋅ a p+n ⋅ vLO
p
=
vLO =VP ⋅ cos (ω P ⋅t )


( p + n)!
= ∑
p=0 p!
⋅ a p + n ⋅ vPp ⋅ cos p (ω P ⋅ t ) (5.35)

From Expression (5.35) we can express the output characteristic i(v) of Figure 5.14 by the
equation:

 VSn ∞
( p + n)! 
i(t ) = ∑  n! ⋅ cosn (ω S ⋅ t ) ⋅ ∑
p!
⋅ a p+n ⋅ VPp ⋅ cos p (ω P ⋅ t ) =

n=0  p=0

∞ ∞
 ( p + n)! n 
= ∑ ∑ 
n=0 p=0 p! ⋅ n!
⋅ VS ⋅ VPp ⋅ a p+n ⋅ cosn (ω S ⋅ t ) ⋅ cos p (ω P ⋅ t )

(5.36)

Taking into account the mathematical development of the cosk(x) given in Appendix I, it is
possible to write the current expression as:

∞ ∞
( p + n)! n  1 C n! 
i(t ) = ∑∑ ⋅ VS ⋅ VPp ⋅ a p+n ⋅  n −1 ⋅ ∑ ⋅ cos (n − 2c) ⋅ ω S ⋅ t + bn   ⋅
n = 0 p = 0 p! ⋅ n!  2  c = 0 (n − c)! ⋅ c!  

 1  D p! 
⋅  p −1 ⋅ ∑ ⋅ cos ( p − 2d ) ⋅ ω p ⋅ t + bp   (5.37)
 2  d = 0 ( p − d )! ⋅ d!  

where:

n − 2 1 n!
n≠0
 2 ⋅ (n/ 2!)2 for n even n≠0
 for n even and and
C= 2 and bn = 
n −1
 for n odd 0
 2  for n odd

p − 2 1
 2 for p even and p≠0 p!
 2 ⋅ ( p/ 2!)2 for p even and p≠0
D= and bp = 
 p − 1 for p odd 0
 2  for p odd
Mixers: Theory and Design 325

Operating on Equation (5.37), we can write:

∞ ∞
( p + n)! n a
i(t ) = ∑∑ n=0 p = 0 p! ⋅ n!
⋅ VS ⋅ VPp ⋅ nn++pp−2 ⋅
2

 C D n! p! 1
⋅ ∑ ∑ ⋅ ⋅ ⋅ cos[(α ⋅ ω S ± β ⋅ ω P ) ⋅ t ] + (5.38)
 c = 0 d = 0 (n − c)! ⋅ c! ( p − d )! ⋅ d! 2
D
p! C
n! 
bn ⋅ ∑ ⋅ cos(β ⋅ ω P ⋅ t ) + bp ⋅ ∑ ⋅ cos(α ⋅ ω S ⋅ t ) + bn ⋅ bp 
d=0 ( p − d )! ⋅ d! c=0 ( n − c )! ⋅ c! 

where α = n − 2c and β = p − 2d and


1
cos(α · ω S · t) · cos(β · ω P · t) = · [cos(α · ω S + β · ω P) · t + cos(α · ω S − β · ω P) · t] =
2
1
= · cos(α · ωS ± β · ωP) · t
2
Observing Equation (5.38) we can obtain the following conclusions:

(a) α and β are equal to or greater than zero.


(b) α ≠ 0 and β ≠ 0 at the first addend.
(c) α = 0 and β ≠ 0 at the second addend.
(d) α ≠ 0 and β = 0 at the third addend.
(e) The fourth addend bn · bp is the term where α = 0 and β = 0.

Taking into account the expressions of bn and bp, given by Appendix I, the Equation (5.38)
can be written as:

∞ ∞ C D
( p + n)! an + p
i(t ) = ∑ ∑ ∑ ∑ (n − c)! ⋅ c! ⋅ ( p − d )! ⋅ d! ⋅ V
n=0 p=0 c=0 d=0
n
S ⋅ VPp ⋅
2 n + p −1
⋅ cos[(α ⋅ ω S ± β ⋅ ω P ) ⋅ t ] +

∞ ∞ D
( p + n)! an+ p
+ ∑ ∑ ∑ (n/2!) ( p − d )! ⋅ d! ⋅ V
n = 0( n even ) p = 0 d = 0
2
n
S ⋅ VPp ⋅
2 n+ p −1
⋅ cos( β ⋅ ω P ⋅ t ) +

∞ ∞ C
( p + n)! an + p
+ ∑ ∑ ∑ ( p/2!) (n − c)! ⋅ c! ⋅ V
n = 0 p = 0( p even ) c = 0
2
n
S ⋅ VPp ⋅
2 n + p −1
⋅ cos(α ⋅ ω S ⋅ t ) +

∞ ∞
( p + n)! a
+ ∑ ∑
n = 0( n even ) p = 0( p even ) ( p / 2!) 2
⋅ ( n / 2!) 2
⋅ VSn ⋅ VPp ⋅ nn ++ pp
2
(5.39)

∞ ∞
where the change ∑α
k =0
k (k = 0, 2, 4, . . . ) = ∑α
k =0
2k (k = 0, 1, 2, . . . ) has been made in the

last three terms of (5.39).


326 Microwave Devices, Circuits and Subsystems for Communications Engineering

Next we will perform a variable change in the different terms of expression (5.40):

α = n − 2c = x ⇒ n = 2c + x
(5.40)
β = p − 2d = y ⇒ p = 2d + y

First term: x ≠ 0 and y ≠ 0


∞ ∞
(2c + 2d + x + y)! V S2c+ x ⋅ V P2 d + y
Ixy = ∑ ∑ ( x + c)! ⋅ c! ⋅ (d + y)! ⋅ d! ⋅
c=0 d =0 2 2c+2 d + x+ y−1
⋅ a2c+2 d + x+ y ⋅ cos( x ⋅ ω S ± y ⋅ ω P ) ⋅ t

(5.41)

Second term: x = 0 and y ≠ 0


∞ ∞
(2n + 2d + y)! V S2n ⋅ V P2 d + y
I0 y = ∑ ∑ (n!)
n=0 d =0
2

⋅ (d + y)! ⋅ d! 2 2 n +2 d + y−1
⋅ a2 n+2 d + y ⋅ cos( y ⋅ ω P ⋅ t ) (5.42)

The first summation is a complete sweep of the variable n and therefore, we can do n ≡ c
without losing the generalization capabilities of Equation (5.42). In this case I0y ≡ Ixy | x=0
obtained from the first term (Eq. (5.41)).
Third term: x ≠ 0 and y = 0. This term is obtained like the second one. In this case
Ix0 ≡ Ixy | y=0 obtained from the first term (Eq. (5.41)).
To calculate the last (DC) term we perform n = c and p = d, so:
∞ ∞
(2c + 2d )! V S2c ⋅ V P2 d
I00 = ∑ ∑ (c!)
c= 0 p = 0
2
⋅ (d!)2
⋅ 2 c +2 d ⋅ a 2 c +2 d
2
(5.43)

1
We can observe that I00 = ⋅ Ixy x=0 .
2 y=0

The frequency components are given by

(n − 2c) · ω S ± (p − 2d) · ω P ⇒ x · ω S ± y · ω P

where

 x = 0, 1, 2 . . .
 and (x, y) cannot be zero simultaneously.
 y = 0, 1, 2 . . .

The harmonic content and intermodulation products are:

ωS ωP ωS ± ωP 2ωS ± ωP ...
2ωS 2ωP ωS ± 2ωP 2ωS ± 2ωP ...
3ωS 3ωP ωS ± 3ωP 2ωS ± 3ωP ...
·· ·· ·· ··
· · · ·
Mixers: Theory and Design 327

5.4.2 Application to the Schottky-Barrier Diode


In the diode of Figure 5.3, we will only consider the current source i(v) which is given by:
i(v) = ISS · [eα ( v+V0 ) − 1] (5.44)
where α is a diode parameter, v is time-varying voltage, V0 is the DC voltage and ISS is the
saturation current. Taking into account the basic non-linear characteristic given in Equation
(5.27), we will deduce the relationship between Equations (5.44) and (5.27). Developing
Equation (5.44) in a Taylor series around V0, we have:

∂i 1 ∂ 2i
i(v) = i(V0 ) + ⋅v + ⋅ ⋅ v2 + . . . (5.45)
∂v v= 0 2! ∂v 2 v= 0

Using Equation (5.44), Equation (5.45) can be written as:

α α2 2
i(v) = IDC + ISS ⋅ eα ⋅V0 ⋅ ⋅ v + ISS ⋅ eα ⋅V0 ⋅ ⋅v +... (5.46)
1! 2!
and comparing Equations (5.27) and (5.46), we can express the coefficients as:

 αk
ak = I SS ⋅ eα ⋅ V0
⋅ for k≠0
 k! (5.47)
a = a = I = I ⋅ [eα ⋅V0 − 1] k=0
 k 0 DC SS for

If we perform the variable change ak = a2c+2d+x+y where k = 2c + 2d + x + y, we can write


Equation (5.41) as:
∞ ∞
(2c + 2d + x + y)! α ( 2 c +2 d + x + y )
Ixy = ISS ⋅ eα ⋅V0 ∑ ∑ ⋅ ⋅ (5.48)
c = 0 d = 0 ( x + c)! ⋅ c! ⋅ ( d + y)! ⋅ d! (2c + 2 d + x + y)!

V S2c+ x ⋅ V P2 d + y
⋅ ⋅ cos( x ⋅ ω S ± y ⋅ ω P ) ⋅ t
2 2c+2 d + x+ y−1
Taking into account that the expressions:
∞ ∞
(α ⋅ VS / 2)2c+ x (α ⋅ VP / 2)2 d + y
Ix (α ⋅ VS ) ≡ ∑
c=0 (c + x )! ⋅ c!
Iy (α ⋅ VP ) ≡ ∑d = 0 ( d + y)! ⋅ d!

are the first class modified Bessel functions (Appendix II), Equation (5.48) can be expressed by:
Ixy = 2 · ISS · eα ·V0 · Ix(α · VS) · I(α · VP) · cos(x · ωS ± y · ωP) · t (5.49)

5.4.3 Intermodulation Power


Figure 5.15 shows a simplified scheme of a single diode mixer. The power Pxy of the
intermodulation product (xy) can be expressed by:
Pxy = Ri · Ixy2 (5.50)
328 Microwave Devices, Circuits and Subsystems for Communications Engineering

V0
IF
Ri
R0 RS

B i B
LO P v P RF
ωP F F
ωS

Figure 5.15 Simplified mixer scheme

where Ixy2 is the root mean square value given by Equation (5.51).

1
Ixy2 = · [2 · ISS · eα ·V0 · Ix(α · VS) · Iy(α · VP)]2 (5.51)
2

Taking into account Equations (5.50) and (5.51), the intermodulation power can be
written as:

Pxy = 2 · Ri · (ISS · eα ·V0)2 · I 2x (α · VS) · I 2y (α · VP) (5.52)

The normal situation of the mixers is that the RF power is very small, in this case α · VS Ⰶ 1
and Ix(α · VS) can be approximated by the first term of the Bessel function (Appendix II),
in this case:

α x ⋅ VSx
Ix (α ⋅ VS ) ≈ (5.53)
2 x ⋅ x!

Therefore, Equation (5.52) can be written as:

α 2⋅x ⋅ VS2⋅x 2
Pxy = 2 · Ri · (ISS · eα ·V0)2 · · I y (α · VP) (5.54)
2 2⋅x ⋅ ( x!)2

where x and y are the RF and LO harmonics respectively and the harmonic frequency is
x · ω S ± y · ω P.
Equation (5.54) is only valid given sinusoidal voltage at the diode and given that Pxy
decreases when the order of x, y grows independently of VS and VP. On the other hand, the
mixer characteristics are a function of the LO amplitude.
As an example, supposing a sinusoidal voltage at the diode for a given LO value, the
harmonic power from Equation (5.54) can be written as:

2 x −1
 α ⋅ ISS ⋅ Ri ⋅ eα ⋅V0 ⋅ Iy (α ⋅ VP )  2 ⋅ x −2  i 
R
Pxy =   ⋅α ⋅
 2
⋅ P1x (5.55)
 x! 
Mixers: Theory and Design 329

Pxy (dBm)

-0 ωS-ωP
2ωS-ωP
-50

-100
-100 -50 -0 P1 a ωS (dBm)

Figure 5.16 Example input power versus output power for two intermodulation
products of a mixer

where P1 is the input power of the RF frequency given by:


x
 1 V 2
P1x =  ⋅ S  (5.56)
 2 Ri 

Where VS is the amplitude of the RF signal. Figure 5.16 shows the harmonic power behaviour
of Equation (5.55).

5.4.4 Linear Approximation


Taking into account the characteristic of Expression (5.27) and vLO(t) and vS(t) given by
(5.28) and (5.29), the input signal in the circuit of Figure 5.14 is given by Equation (5.30).
If we suppose that Vp Ⰶ VS (Figure 5.17), the first order of the Taylor series around vLO of the
characteristic i(v), for a DC voltage V0 (operating point), can be written as

1  di 
i(v) = i(vLO + vS) ≈ i(vLO) + ⋅ · vS = i(vLO) + VS · cos(ωS · t) · g(1)(vLO) (5.57)
1!  dv  vS = 0

i(v)

i(t)
I0
V0 V(t)
VS

VLO

Figure 5.17 Linear approximation


330 Microwave Devices, Circuits and Subsystems for Communications Engineering

where

i(vLO) = ∑V
k= 0
P
k
· ak · cosk(ωP · t) (5.58a)


k!
g(1)(vLO) = ∑ (k − 1)! · a
k =1
k · VPk−1 · cosk−1(ωP · t) (5.58b)

Putting Equations (5.58) into (5.57), we can write:



∞ k! 
i( v ) = ∑α
k =0
k ⋅ VPk ⋅ cosk (ω P ⋅ t ) + VS ⋅ cos(ω S ⋅ t ) ⋅ ∑
 k =1 (k − 1)!
⋅ ak ⋅ VPk−1 ⋅ cosk−1(ω P ⋅ t )

(5.59)

It is important to observe that the DC contribution to the current I0 is not written in Expression
(5.59). I0 is the value of the current at the operating point V0. Developing Equation (5.59),
we can obtain:
i(t) = a0 + a1 · VP · cos(ωP · t) + a1 · VS · cos(ωS · t) + 2 · a2 · VS · VP · cos(ωS · t) · cos(ωP · t) +
∞ ∞
(n + 1)!
+ ∑ an · V Pn · cosn(ωP · t) + VS · cos(ωS · t) ·
n= 2

n =2 n!
· an+1 · V Pn · cosn(ωP · t) (5.60)

Looking at Appendix I, the penultimate term of Equation (5.60) can be expressed as follows:
∞ ∞
 1 C
n! 
∑a
n =2
n ⋅ VPn ⋅ cosn (ω P ⋅ t ) = ∑a
n =2
⋅ VPn ⋅  n−1 ⋅ ∑
n
 2 c = 0 (n − c)! ⋅ c!
⋅ cos[(n − 2 ⋅ c) ⋅ ω P ⋅ t  +


1 n! n − 2
 2 for n even
+ n ⋅
2 n where C = n − 1 (5.61)
!
2 n even  for n odd
 2
and the last term of Equation (5.60) will be:

(n + 1)!

n=2 n!
⋅ an +1 ⋅ VPn ⋅ cosn (ω P ⋅ t ) =

∞ C
(n + 1)! n! 1
= ∑∑a
n=2 c=0
n+1 ⋅ VPn ⋅
n!
⋅ ⋅ n −1 ⋅ cos[(n − 2 ⋅ c) ⋅ ω P ⋅ t ] +
(n − c)! ⋅ c! 2

(n + 1)! 1 n! C
+ ∑
n = 2 ( n even )
an +1 ⋅ VPn ⋅
n!
⋅ n ⋅
2 (n/ 2)!2
∑ . . . cos[(n − 2 ⋅ c) ⋅ ω
c=0
P ⋅ t]

n − 2
 2 for n even
where C= (5.62)
 n − 1 for n odd
 2
Mixers: Theory and Design 331

Considering Equations (5.61) and (5.62), Equation (5.60) can be written as:

i(t) = a0 + a1 · VP · cos(ωP · t) + a1 · VS · cos(ωS · t) + 2 · a2 · VS · VP · cos(ωS · t) · cos(ωP · t) +

∞  ∞ 
n! 1  (n + 1)! 1 n
+ ∑ ⋅
2n
an ⋅ VP
n
+
n∑ 2

2n
an +1 ⋅ VP ⋅ VS ⋅ cos(ω S ⋅ t ) +

(5.63)
n = 2 (n / 2)! = 2 ( n / 2) !
n even  n even 
∞ C
n! 1
+ ∑∑a n ⋅ VPn ⋅ ⋅
(n − c)! ⋅ c! 2 n−1
⋅ cos[(n − 2 ⋅ c) ⋅ ω P ⋅ t ] + ← Harmonics of ω P
n =2 c = 0

∞ C
(n + 1)! 1
+ ∑∑a n +1 ⋅ VPn ⋅ VS ⋅ ⋅
(n − c)! ⋅ c! 2 n
⋅ cos[(n − 2 ⋅ c) ⋅ ω P ⋅ t ± ω S ⋅ t ] ← Intermodulation
n =2 c = 0

Harmonics of the local oscillator [(n − 2c)ωP]:

n = 2 c = 0 2 · ωP
n = 3 c = 0, 1 3 · ωP ωP
n = 4 c = 0, 1 4 · ωP 2 · ωP
n = 5 c = 0, 1, 2 5 · ωP 3 · ωP ωP
n = 6 c = 0, 1, 2 6 · ωP 4 · ωP 2 · ωP

Intermodulation products [(n − 2c)ω P ± ω S]:

n=2 c=0 ωS ± 2 · ωP
n = 3 c = 0, 1 ωS ± 3 · ωP ωS ± ωP
n = 4 c = 0, 1 ωS ± 4 · ωP ωS ± 2 · ωP
n = 5 c = 0, 1, 2 ωS ± 5 · ωP ωS ± 3 · ωP ωS ± ωP
n = 6 c = 0, 1, 2 ωS ± 6 · ωP ωS ± 4 · ωP ωS ± 2 · ωP

where ω FI = ω S − ω P. The spectral lines of the current i(t) are given in Figure 5.18. The
distance between the harmonics of ω P and the intermodulation products is ω FI. These
intermodulation products are symmetric and occur at frequencies given by:

ω n = ω FI + n · ω P where n = . . . −3, −2, −1, 0, 1, 2, 3, . . .

5.5 Diode Mixer Theory


We will start from the non-linear equivalent circuit of Figure 5.3 and from the dynamic
model of Figure 5.9. The non-linear equations are given by Equations (5.1) and (5.2).
Taking into account Equation (5.1), the associated incremental conductance g(v) at each v
point can be expressed by:

di q⋅v q⋅v
q⋅v
g(v) = = ISS ⋅ ⋅ e n⋅k ⋅t ≈ ⋅ i( v ) (5.64)
dv v n⋅k ⋅t n⋅k ⋅t
332 Microwave Devices, Circuits and Subsystems for Communications Engineering

I(t)

ωFI
DC

ωFI ω-1 ωP ω1 ω-2 2ωP ω2 ω-3 3ωP ω3 ω


Figure 5.18 Harmonic representation of the i(t) nonlinear current

Also, we have an incremental charge value, associated with the non-linear capacitor, in each
v point. The alternating current through the capacitor can be written as:

dQ dQ dv dv
ic (t ) = = ⋅ = C(v) ⋅ (5.65)
dt dv v dt dt

We suppose quasi-linear excitation (RF amplitude Ⰶ LO amplitude). In this case, the diode
can be considered to be non-linear with respect to LO signal and quasi-linear with respect to
RF signal. Under LO excitation, the non-linear voltage at the diode terminals of Figure 5.3
is given by:

[ ] dv
q⋅v
vd = v(t ) + RS ⋅ ISS ⋅ e n ⋅ k ⋅ T − 1 + RS ⋅ C(v) ⋅ (Large signal) (5.66)
dt

and the non-linear current:

[ ] dv
q⋅v
i(t ) = ISS ⋅ e n ⋅ k ⋅ T − 1 + C(v) ⋅ (Large signal) (5.67)
dt

Therefore, the way to solve the single-diode mixer problem is:

1. To solve the non-linear circuit for LO excitation and to obtain all the harmonics of ω P
(Figure 5.19).
2. To do linear superposition of the RF signal (Figure 5.19).

5.5.1 Linear Analysis: Conversion Matrices


We will apply two signals at the diode terminals: the RF signal, whose frequency is ω S,
and the LO signal, whose frequency is ω P. We will suppose that the amplitude VS of the ω S
signal is much less than the amplitude VP of ω P. That implies that VS is a small perturbation
around VP. In this case, C(v) and g(v) from Equations (5.5) and (5.6) and Figure 5.3, vary
harmonically under LO excitation.
Mixers: Theory and Design 333

R0 ωP
ωIF

Id Intermod.
Solution
Matching
Non Linear Linear
vd Linear Intermod. LO RF
Network Pumping Superposition

ωS RS

Figure 5.19 Single-diode mixer solution

If we know

id (t) = Ido + ∑ Re( I


n
dn ⋅ e j⋅n⋅ω P⋅t )
(5.68)
vd (t) = Vdo + ∑ Re(V
n
dn ⋅ e j⋅n⋅ω P⋅t )

we can find v(t) (Figure 5.3) and in this case we know the Fourier series of:

C(v) ⇔ C(t)
g(v) ⇔ g(t)

5.5.1.1 Conversion matrix of a non-linear resistance/conductance


The excitation i(t) is a harmonic temporal function (LO) therefore the response v = f(i) is a
dv
harmonic temporal function (Figure 5.20). In this case, r(i) = = r(t) is also a harmonic
dt
temporal function that depends on LO and it is possible to develop it as a Fourier series:

i i
V=f(i)
r(i)=dv/dt
V=f(i) r(i)=dv/dt
V0
I0
i i

Figure 5.20 Non-linear resistance under LO excitation


334 Microwave Devices, Circuits and Subsystems for Communications Engineering


r(t ) = ∑R
−∞
n ⋅ e j⋅n⋅ω P⋅t (5.69)

where Rn = Rn* are the Fourier coefficients.


On the other hand, when two tones are applied in a non-linear element, intermodulation
frequencies appear (linear approximation), as we saw in Equation (5.63). The intermodulation
frequencies are given by n · ωS ± m · ω P or ω n = ω FI + n · ωP with ω FI = ω S − ω P. The
generic form of the voltages and currents of the intermodulation products is given by the
sum of all of their components:

I(t ) = Re ∑V
n = −∞
n ⋅ e j ⋅(ω FI +n ⋅ω P )⋅t

H(t ) = Re ∑I
n = −∞
n ⋅ e j ⋅(ω FI +n ⋅ω P )⋅t (5.70)

where I(t) and H(t) are very small signals. In this case, we can express Ohm’s Law as:
I(t) = r(t) · H(t) (5.71)

and developing Equation (5.71) by Equation (5.70):


l =∞ m =∞ n =∞

∑V ⋅ e
l = −∞
l
j ⋅ (ω FI + l ⋅ω P ) ⋅ t
= ∑R
m= −∞
m ⋅ e j ⋅ m ⋅ω P ⋅ t ⋅ ∑I
n = −∞
n ⋅ e j ⋅(ω FI +n ⋅ω P )⋅t =

m = ∞ n =∞
= ∑ ∑R
m = −∞n = −∞
m ⋅ In ⋅ e j ⋅(ω FI +( n⋅+ m )⋅ω P)⋅t (5.72)

Equating both parts of Equation (5.72), we obtain the conversion matrix of the non-linear
resistor r(t):

 C − N   E0 E−1 L E− N L E−2 N +1 E−2 N   C − N 


     
F− N +1   E1 E0 L E− N +1 L E−2 N +2 E−2 N +1  C − N +1 
     
M  M M M  M 
     
F0  =  EN EN −1 L E0 L E− N +1 EN  ⋅  C 0  (5.73)
     
M  M M M  M 
     
F   E E2 N −2 L EN −1 L E0 E1   C N −1 
 N −1   2 N −1   
     
FN   E2 N E2 N −1 L EN L E1 E0   C N 

The matricial Equation (5.73) can be written as:


[F] = [E] · [C] (5.74)
where [E] is called the conversion matrix.
Mixers: Theory and Design 335

The intermodulation product response (voltage matrix) is a function of the excitation and
the known conversion matrix. The conversion matrix [E] is a function of the circuit and
the LO.
We have an analogous situation for calculating a non-linear conductance as in the diode
case of Figure 5.3:

g(t ) = ∑G
−∞
n ⋅ e j ⋅ n ⋅ω P ⋅ t (5.75)

and the matricial equation is:

[C] = [B] · [F] (5.76)

5.5.1.2 Conversion matrix of a non-linear capacitance


The excitation v(t) is a harmonic temporal function (LO) and the response Q(t) = f[(v(t)] is
also a harmonic temporal function (Figure 5.21). The capacitance is given by:

dQ
C(v) = (5.77)
dv

and it will also be a harmonic temporal function. Developing this as a Fourier series, we
have:

C(t ) = ∑C
n =−∞
n ⋅ e j ⋅ n ⋅ω P ⋅ t (5.78)

where An are the Fourier coefficients.


When we introduce a small signal ωS along with the LO excitation ω P, intermodulation
frequencies appear as in the resistance case where ω n = ω FI + n · ω P with ω FI = ω S − ω P.
Now we can apply the linear approximation:

dI
H (t ) = C(t ) ⋅ (5.79)
dt

I(t)
C(v)

V(t) V0
C(v)
V(t)

C(v)=dQ/dv

Figure 5.21 Non-linear capacitance under LO excitation


336 Microwave Devices, Circuits and Subsystems for Communications Engineering

The generic form of the voltages and currents of the intermodulation products is given by
Equation (5.70) and therefore, we can write:
m =∞ k =∞ n =∞

∑ Im ⋅ e j ⋅(ω FI +m⋅ω P)⋅t =


m =−∞
∑ Ck ⋅ e j⋅k⋅ω P⋅t ⋅
k =−∞
∑ j ⋅ (ω
n =−∞
FI + n ⋅ ω P ) ⋅ Vn ⋅ e j ⋅(ω FI +n ⋅ω P)⋅t (5.80)

Equating both parts of Equation (5.80), we obtain the matricial Equation (5.81):

 C− N  F− N 
  A0 ⋅ ω − N A−1 ⋅ ω − N +1 L A−2 N ⋅ ω N   
 C− N +1    F− N +1 
  A1 ⋅ ω − N A0 ⋅ ω − N +1 L A−2 N +1 ⋅ ω N   
M    M 
  M M M   
 C0  =   ⋅ F0  (5.81)
  AN ⋅ ω − N AN −1 ⋅ ω − N +1 L A− N ⋅ ω N   
M    M 
  M M M   
 C N −1    FN −1 
  A2 N ⋅ ω − N A2 N −1 ⋅ ω − N +1 L A0 ⋅ ω N   
 C N  FN 

Each current response sub-index is obtained from the capacitor sub-index plus the voltage
sub-index because the sub-index of the frequency is the same as that of the voltage.
Ik = Cj · ωn · Vn j+n=k (5.82)
The matricial Equation (5.81) can be written as:
[C ] = j · [A] · [g] · [F] (5.83)

where the matricial current response is a function of the RF excitation and matrix [A] is a
function of the circuit and the LO signal. The product [A] · [g] is called the conversion
matrix of the a non-linear capacitance. The matrices [A] and [g] are given by:

A0 A−1 L A−2 N 


  ω − N 0 0 L 0 
A1 A0 L A−2 N +1  
  0 ω − N +1 0 L 0 
M M M   
[C] =   [Ω] = 0 0 ω − N +2 L 0 
AN AN −1 L A− N   
  M M M M 
M M M   
  0 0 0 L ω N 
A2 N A2 N −1 L A0 

5.5.1.3 Conversion matrix of a linear resistance


In the case of the linear resistance of the diode equivalent circuit of the Figure 5.3, the
matricial equation voltage/current can be written as:
[ VS ] = [ RS ] · [ Id ] (5.84)
Mixers: Theory and Design 337

where [ VS ] is the matrix of the difference of potential at RS terminals, [ Id ] is the matrix of


the total current trough the diode and the [ RS ] matrix is given by:

 RS 0 0 L 0 
 
0 RS 0 L 0 
 
[ RS ] =  0 0 RS L 0  (5.85)
 
M M M M 
 
0 0 0 L RS 

The same criterion we will apply for any linear impedance.

5.5.1.4 Conversion matrix of the complete diode


Taking into account the Schottky diode model of Figure 5.3, the voltage/current matricial
relationship is:

[ Vd ] = [ Zd ] · [ Id ] (5.86)

where [ Vd ] and [ Id ] are the matrices of the total voltages and currents at the external diode
terminals and [ Zd ] is the complete impedance conversion matrix given by Equation (5.87)
and it is only valid for the intermodulation product frequencies.

[ Zd ] = [ RS ] + ([B] + j · [A] · [g])−1 (5.87)

5.5.1.5 Conversion matrix of a mixer circuit


The complete pumped diode behaves as a frequencial multiport network as we can see in
Figure 5.22.
Zen is the impedance that the diode sees at each frequency ωn. [ Zd ] is an exclusive
function of the diode parameters and the power at the frequency of the local oscillator, for a
given LO. Therefore, our interest is the relationship between RF and IF ports. For this
purpose, we open each port as indicated in Figure 5.23.

I1
Ze0
ωS V1
DIODE ωFI
Zd
Ze-1 Ze2
2N+1
Ports Intermod
Ze-2

Figure 5.22 Diode mixer


338 Microwave Devices, Circuits and Subsystems for Communications Engineering

Z e1 Ze0

DIODE
Ze-1 Zd Ze2

[Z ] matrix
a
Figure 5.23 d

I0 I1

y 00 y 01
V0 V1
y10 y11

Figure 5.24 Relationship between IF (0) and RF (1) ports

The process to obtain the relationship between RF and IF is the following:

1. Sum the diagonal linear matrix [ Zen ] with the conversion matrix [ Zd ]:

[ Zd ]a = [ Zen ] + [ Zd ] (5.88)

2. Calculate the inverse matrix of [ Zd ]a : [ Y ] = ([ Zd ]a)−1.


3. Null all unwanted intermodulation frequencies. In this case we have:

 I0   y00 y01  V0 


 = ⋅  (5.89)
 I1   y10 y11  V1 
where 0 corresponds to the IF port and 1 corresponds to the RF port (Figure 5.24).

5.5.1.6 Conversion gain and input/output impedances


From the circuit of Figure 5.24, we can obtain the circuit of Figure 5.25 introducing the RF
generator and drawing in explicit form the RF and IF impedances.
Taking into account Figure 5.25, we can define the transference power gain, or gain
conversion, as:

PIF
GC = (5.90)
PRF
Mixers: Theory and Design 339

I1 I0

Ze1 CONVERSION
RF V1 Ze0 IF
NETWORK
Zin Zout

Figure 5.25 Mixer circuit

where PIF is the dissipated power at the IF port and PRF is the available power at the RF
generator.
From Equation (5.89) when V0 = 0, we can express the matrix parameters as:

I0 I1  I0 = y01 ⋅ V1
y01 = y11 = ⇔  (5.91)
V1 V0 =0
V1 V0 =0  I1 = y11 ⋅ V1
In this case, the dissipated power PIF will be:

1 2 1
PIF = · I0 · Re(Ze0) = · |y01|2 · V 21 · Re(Ze0) (5.92)
2 2

and the available power PRF :


1 V 12
PRF = ⋅ (5.93)
8 Re( Ze1 )

Taking into account Equations (5.90), (5.91) and (5.92), the conversion gain, supposing the
circuit is coupled without losses, is:

GC = 4 · |y01 |2 · Re(Ze1) · Re(Ze0) (5.94)

From Figures 5.24 and 5.25, we can deduce the expressions of the input and output impedance:

1 1
Zin = − Ze1 Zout = − Ze 0 (5.95)
y11 y00

If we generalise the gain conversion and the impedances at any intermodulation products,
we obtain the following expressions:

GCmn = 4 · | ymn |2 · Re(Zen) · Re(Zem) (5.96)

1 1
Zin = − Zen Zout = − Zem (5.97)
ynn ymm

5.5.2 Large Signal Analysis: Harmonic Balance Simulation


Under large signal excitation (LO signal) the diode behaviour follows the non-linear
Equations (5.67) and (5.68) and the objective of the large signal analysis is to calculate the
Fourier components V(n · ωP) of the internal voltage v(t). When the components V(n · ωP)
340 Microwave Devices, Circuits and Subsystems for Communications Engineering

Z RF
Id
Matching

vd Linear Z IF

Network

R0 ωP

LO

Figure 5.26 Mixer circuit under LO excitation

Id (t) IL(t)
DIODE R0
I(vj ) + + Linear
Vj (t) LO
Network ωP
C(vj) - VL(t)
Non Linear

Figure 5.27 Linear and non-linear mixer networks under LO excitation

are known, v(t) is known. Taking into account Equations (5.65) and (5.66), we can know the
Fourier components of Equations (5.76) and (5.78) and therefore the complete conversion
matrix (5.87).
The solution of the multi-harmonics id(t) and vd (t) given in Equation (5.69), through the
non-linear device (diode) under large signal excitation can be obtained by harmonic balance
analysis method. This analysis can be implemented in several ways but all of them can be
explained under the same idea:
The mixer circuit under large signal excitation (Figure 5.26) can be divided in two parts:
the linear and non-linear circuits as we can observe in Figure 5.27. I(vj) and C(vj) are the
junction current source and junction capacitance respectively, which depend on the junction
voltage vj. Id (t) is the external non-linear current of the diode and VL(t) and IL(t) are the
voltage and current of the linear circuit.
The basic idea of the harmonic balance method is quite simple. For a given local oscillator
and an initial condition of the linear network, we can calculate the linear voltage and current
VL(t) and IL(t). The voltage VL(t) must be equal to the junction voltage Vj (t), with this voltage
we can calculate the non-linear current Id (t) and this current is compared with IL(t). We will
reach the stationary solution when the currents have the same amplitude and a phase differ-
ence of 180 degrees.
Normally, the linear network is calculated in the frequency domain and the non-linear one
is calculated in the time domain. A good Fourier transforming algorithm is necessary to
change from the temporal to the frequency domains and the frequency to time domain. The
number of harmonics that we take into account should be chosen carefully, if the number of
harmonics is low, the solution will not be correct, but if the number is high, the calculation
time can be very high.
Mixers: Theory and Design 341

Finding the stationary solution is an optimisation problem. It is necessary to define an


appropriate error function and to search for the zeros of this function by an optimisation
algorithm. The advantage of optimisation is that a great number of optimisation subroutines
are already available in computer mathematics libraries.
One of the commonly used optimisation techniques is the Newton’s method or Newton-
Raphson method. It is an iterative method for finding the zeros of the error function and it
needs to know the gradient of the error function. It is an efficient method when the derivatives
are easily evaluated and we can make a good initial estimation of the solution.
There are other optimisation methods (relaxation and reflection algorithms, for instance)
but it is not the objective of this chapter present a study of them. In any case, all the modern
non-linear and large signal simulators have implemented these methods.

5.6 FET Mixers


Any device used in a mixer must have a strong non-linearity, low noise, low distortion and
an adequate frequency response. Traditionally, the Schottky diode has been the most used
non-linear device for mixers. The field-effect-transistors (FET) have become very popular in
the past few years as mixer devices. The main reason is due to the great advantage that
GaAs’ microwave monolithic integrated circuit can produce. FET transistors are more suitable
than diodes in this technology, since non-planar structures as phase shift circuits are required
in the balanced mixers. Nevertheless, the most important characteristic of the FET mixers is
that they can exhibit conversion gain well into the millimetre-wave region.
FET mixers can be designed with good noise performance as well as conversion gain, and
lower LO power is required, if compared to that of diode mixers. Since FETs are available
as dual-gate devices, the LO and RF can be applied to separate ports, improving the isolation
between these ports. Balanced FET mixers are also possible, and they have the same LO
noise rejection and spurious properties as balanced diode mixers.

5.6.1 Single-Ended FET Mixers

5.6.1.1 Simplified analysis of a single-gate FET mixer


The square-law characteristic of a FET can be used in a frequency conversion. Figure 5.28
shows a simplified scheme of a FET mixer where both RF and LO signals are injected
through the gate of the transistor.
The drain current can be expressed as:
2
 V 
Id = Idss 1 − gs  (5.98)
 Vp 

∂Id
The transconductance is defined as: gm = and substituting Id into the last expression:
∂Vgs

−2 Idss  Vgs 
gm = 1 − V  (5.99)
Vp  p 
342 Microwave Devices, Circuits and Subsystems for Communications Engineering

VDD

RD IF

LO

RF

VSS

Figure 5.28 FET mixer with RF and LO both applied at the gate

However, Vgs = Vg + VLO cos ω LO t (VLO Ⰷ VRF) where VLO is the local oscillator voltage amplitude,
VRF is the radiofrequency voltage amplitude and Vg is the gate-source bias voltage. The
transconductance is now:

 −2 Idss   V + VLO cos ω LO t   −2 Idss   V   −2 Idss 


gm =  1− g  =   1 − g +  VLO cos ω LO t (5.100)
 Vp  
  Vp   Vp   Vp   V p2 

The small-signal drain current is:

 −2 Idss   V  −2 IdssVLOVRF 
id (t ) = gm (t )VRF (t ) =    1 − g  VRF cos ω RF t +   cos ω LO t cos RF t (5.101)
 Vp   Vp   V p2 

The large-signal LO modulates [1] the transconductance (gm) of the device and when a
small-signal RF is applied simultaneously, the small-signal drain current is proportional to
their product (VLO VRF).
As the time-varying transconductance is the main contributor to mixing, these mixers are
called transconductance mixers, and the mixing products attributable to parametric ‘pumping’
of the gate-source capacitance, gate-drain capacitance and drain-source resistance can be
considered negligible. In order to get the maximum conversion gain is important to maximise
the range of the MESFET’s transconductance variation and, in particular, the magnitude of
the fundamental frequency component of the transconductance. To maximise the magnitude
of the fundamental component of gm, the device must be biased close to the pinch-off value,
Vp, and must remain in the saturation region throughout the LO cycle. The best way to
ensure this can be achieved by short-circuiting the drain terminal at the LO frequency and
all LO harmonics.
The input matching circuit must match the RF source to the MESFET’s gate, and short-
circuit at the IF frequency to avoid the amplification of any input noise at this frequency.
A good output matching circuit is important since an inadequate output network can cause
instablility. As well, this network must be a good IF filter in order to get good isolation between
the ports. In many cases it is possible for the IF circuit to provide both impedance transforma-
tion and filtering functions via a single structure, which always minimises circuit loss.
Mixers: Theory and Design 343

Figure 5.29 Large-signal non-linear MESFET equivalent circuit

Figure 5.30 Large-signal equivalent circuit of the FET mixer

5.6.1.2 Large-signal and small-signal analysis of single-gate FET mixers


The FET mixer can be analysed via the large-signal procedure [2] similar to that described in
Section 5.3. The large-signal non-linear MESFET equivalent circuit is shown in Figure 5.29.
If the transistor is biased in its saturation region through the LO cycle, the non-linearities of
Id and Cgs can be simplified a lot, and Cgd can be considered as a linear element. In this case,
the large signal equivalent circuit of the FET mixer is shown in Figure 5.30, where ZS (nωp)
and ZL(nωp) are the embedding impedances of the source and load respectively, and ωp is the
LO frequency.
The most used method to perform large-signal analysis of FET mixers is the harmonic-
balance method. This method calculates only the steady-state solution for the circuit. The
non-linear circuit is divided into linear and non-linear subcircuits. The linear subcircuit can
be treated as a multiport and described by its y-parameters, s-parameters, or some other multi-
port matrix. The non-linear elements are modelled by their global I/V or Q/V characteristics,
and must be analysed in the time domain.
344 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 5.31 Small-signal equivalent circuit of the FET

The idea of harmonic balance is to find a set of port voltage waveforms (or, alternatively,
the harmonic voltage components) that gives the same currents in both the linear-network
equations and the non-linear-network equations. When that is satisfied, we have the solution.
The aim of a small-signal analysis is to calculate the conversion gain and input and output
impedances of the mixer, using a linear small-signal FET equivalent circuit, as shown in
Figure 5.31.
The mixing takes place in the transistor when the small-signal elements are varied periodic-
ally by a large LO signal, which is applied between the gate and source terminals. For a
GaAs MESFET, the major dependence with the gate bias is produced by the transconductance,
gm. The mixing products produced by the Cgs capacitance and the Ri resistance are considered
negligible. For the gate-pumped mixers the drain resistance, Rds, variation is small, so the time-
averaged value is used. As the main contributor to the frequency conversion is produced by
the variation of the tranconductance, these mixers are called ‘transconductance mixers’.
When a large LO signal is applied between the gate and source terminals, the trans-
conductance becomes in a time-varying function gm(t) with a period equal to that of the LO.
If ωo is the LO frequency:

gm (t ) = ∑g e
k =−∞
k
jkω o t
(5.102)

where:

gk =
1
2Π 冮 g (t)e
0
m
− jkω o t
d (ω ot ) (5.103)

are the Fourier coefficients of the transconductance. Of these coefficients g1 is the most
important, which corresponds to the fundamental component of gm in the frequency domain.
This coefficient is a function of the LO signal amplitude, of the gate bias, and of the shape
of the curve gm /Vgs. The value of g1 is not greater than gmo /A, where gmo is the maximum
value of gm. For the ideal case, when g1 is a step function of the gate voltage:

g1 1
= (5.104)
gm Π
Mixers: Theory and Design 345

Figure 5.32 Small signal equivalent circuit of a FET mixer

Figure 5.32 shows the equivalent circuit of a FET mixer [3], where ω1 corresponds to the RF
frequency and ωo to the LO signal. V1, V2, V3 and I1, I2, I3 are the complex voltage and current
amplitudes of the signal, image and intermediate frequency in the gate circuit, and V4, V5, V6
and I4, I5, I6 the corresponding voltage and current amplitudes of the drain circuit. E1
represents the voltage source for the RF signal, with internal impedance Z1, and the rest of
components are terminated in complex impedances.
The circuit shown in Figure 5.32 can be analysed with the loop equations for each
frequency component. In matrix notation these equations are written as:

[E] = [V] + [Zt][I] = [Zm][I] + [Zt][I] (5.105)

Where:

 E *1 V *1   I *1 
     
0  V2   I2 
     
0  V3   I3 
[ E] =   [V ] =   [ E] =  
V  V *4   I *4 
     
0  V5   I5 
     
0  V6   I6 

and [Zm] and [Zt] are, respectively, the matrices representing the proper mixer and its termi-
nations. They are given by:
346 Microwave Devices, Circuits and Subsystems for Communications Engineering

 Z 11
* 0 0 * 0
Z 14 0 
 
0 Z22 0 0 Z25 0 
 
0 0 Z33 0 0 Z36 
[ Zm ] =  
 0 Z43 * 0
Z 44 0 
 
0 Z52 Z53 0 Z55 0 
 
*
 Z 61 Z52 Z63 0 0 Z66 

 Z *1 0 0 0 0 0 
 
0 Z2 0 0 0 0 
 
0 0 Z3 0 0 0 
[ Zt ] =  
0 0 0 Z *4 0 0 
 
0 0 0 0 Z5 0 
 
0 0 0 0 0 Z6 

The available conversion gain between the RF input (port 1) and the IF output (port 6) is:

| I6 | 2 Re( Z6 )
Gav = (5.106)
| E1 | 2 / 4Re( Z1 )

When the IF frequency is small compared to the input signal frequency, many simplifications
can be made. The conversion gain is maximum when the source and load are conjugately
matched to the FET. In this case the conversion gain is given by:

g12 Rd
Gav, max = (5.107)
4ω 12 A 2 Rin

where Ed is the time-averaged value of Rds, A is the time-averaged value of Cgs, and Rin = Rgm
+ Ri + Rs.

5.6.1.3 Other topologies


The gate-pumped FET mixer is the most used topology and the best known. Nevertheless, in
many applications it is not possible, nor convenient, to apply the LO signal through the gate.
Drain-pumped mixers are frequently used when LO and RF signals are separate in frequency
and it is not possible to design a combiner with sufficient performance. Then, the LO signal
is injected through the drain terminal and the IF signal is extracted at the same port with the
aid of a diplexer.
Figure 5.33 shows the small-signal equivalent circuit for a drain mixer [4]. When a large
signal is applied between two FET terminals, the small signal elements are varied periodic-
ally. The main non-linearities are the transconductance gm and the channel resistance Rds. For
Mixers: Theory and Design 347

Figure 5.33 Small-signal equivalent circuit of a drain mixer

a gate-pumped design and under saturation drain bias, the gm non-linearity is more significant
than that of the Rds. Therefore a time-averaged value is taken for Rds. But for a drain-pumped
mixer, the transconductance and the channel resistance are modulated by the LO signal and
become in a time-varying functions with a period equal to that of the LO. Therefore, the
amplification factor µ = gm Rds becomes in a time-varying function as well.
When a small-signal of frequency ω 1, is applied to the gate-source terminals, both non-
linearities µ(t) and Rds(t) are contributing to the mixing. Only the intermediate frequency
ω 3 = ω LO − ω 1 and the image frequency ω 2 = 2ω LO − ω 1 are considered as generated product
by the mixing. The rest of the mixing products are considered to be eliminated by the filters
Fk, k = 1, . . . 6, which are supposed as ideal band-pass filters. gm and Rds are the only non-
linearities of the circuit. Rgm, Rs, Rdr, Cgs and Cgd are considered constants in the analysis.
If the non-linearities are developed in Fourier series, with µ n and Rdn as coefficients of both
series, the conversion matrix can be obtained (similar to the development in Section 5.6.1.2).
When Cgd is considered zero, the conversion gain is:
2
I
GM = 4 RG RL 6 (5.108)
E1

where RG is the real part of the signal source impedance and RL is the real part of the IF load
impedance. When Z3, Z4 and Z5 have a high value and the input and output circuits are
conjugately matched, the conversion gain is:

| µ1 | 2
GM = (5.109)
4ω C ( Rgm
2
1
2
gs + Rs )( Rdr + Rs + Rdo )

where Rdo is the DC component of Rds.


348 Microwave Devices, Circuits and Subsystems for Communications Engineering

In this last expression, the non-linearity of Rds only appears in the fundamental component
of the amplification factor µ, which differs from the expression of the conversion gain for a
gate-pumped mixer. For gate LO pumping the gain is a function of the fundamental com-
ponent of gm, and a time-averaged value is taken for Rds.
A third way to inject the LO signal is using the source terminal. This topology is less
common than the other two ones, but it is used when LO and RF are very distant in
frequency and can be filtered properly. Normally, RF signal is injected through the gate and
the IF signal is extracted from the drain. Because Cgs is not very small, bad LO/RF isolation
will result if filters are not added in the RF and LO ports. For this reason, LO and RF
bands cannot be close in frequency. For a source-pumped mixer, the transconductance and
the channel resistance become time-varying functions, so both of them must be taken into
account in order to design the mixer properly.
Recently several authors have studied resistive FET mixers. These have conversion losses
and noise figures comparable to those of the diode mixers, but can achieve much better IM
and spurious signal performance.
Mixers are usually the most non-linear devices of a receiver front end. For this reason
the intermodulation (IM) performance is often limited by the mixer, and furthermore, this
device is the only stage that generates spurious signal responses. For some applications,
where broadband behaviour is required, these characteristics may be more limiting than
noise. On the other hand, low intermodulation levels are required in some digital systems
(like OFDM modulation systems), which has led to an improvement in the distortion level
required of the mixers.
Traditionally, mixers have been constructed with a large LO signal and a small RF
signal applied to a non-linear device. The large LO voltage changes the union impedance
between a very small value and nearly an open circuit. This time-varying resistance is
responsible for the mixing, and these mixers are called ‘resistive mixers’. The non-linearity
of the union presents a time-varying resistance, since the slope of the I/V curve is changed
when pumped by the LO signal. If a linear time-varying resistance can be achieved, then
intermodulation-free mixing would be reality.
An ideal linear time-varying resistance cannot be created, but it is possible to find some-
thing close to it, such as the channel of an unbiased GaAs MESFET. The channel resistance
of a cold MESFET (with no drain bias) can change when a signal is applied to the gate
terminal. If the transistor is drain-biased with a very small (or zero) voltage, then the relation
between the drain to source current and the gate-source voltage is non-linear. However,
drain-source current varies almost linearly with drain-source voltage. This last characteristic
permits a very small distortion level compared to that generated by a diode. Since the channel
resistance is non-linear with Vgs, the LO signal must be applied between these two terminals.
The RF signal will be injected through the drain, and now that Rds is almost ‘linear’ with Vds,
very little distortion is generated.
Since the LO signal is modulating the channel conductance, it can be expanded in a
Fourier series:

Gds = go + 2g1 cos(ωLO t) + . . . (5.110)

where ω LO is the LO frequency signal. Taking into account only the first two terms and when
load and source are conjugately matched, the conversion losses are given by [5]:
Mixers: Theory and Design 349

g 02
Ic ≅ (5.111)
g12

In order to minimise the conversion losses, g1 must be as large as possible. Thus, the
transistor will be biased to maximise the fundamental component of the channel conduct-
ance, g1. This bias point corresponds to that where the channel resistance is most sensitive to
bias modulation by the LO. This point is a little above the pinch-off value, where the
channel conductance is most non-linear with Vgs.
Since drain-gate capacitance is greater for non-biased transistors compared with the same
transistor in the saturation region, there is a coupling between these ports. Therefore, LO
signal leakage will appear in the drain, increasing the drain-source voltage and the
intermodulation level generated. A low DC value at drain port can be obtained by shorting
the terminal at LO frequency and its harmonics. Balanced topologies can be used when LO
and RF frequency bands are very close. An example of a double-balanced resistive FET
mixer is shown in Section 5.9.

5.7 Double-Gate FET Mixers


When designing a mixer with a single gate device, the first problem is how to apply both
signals, RF and LO, to the transistor gate. Passive couplers are commonly used in conven-
tional hybrid technology. Nevertheless, for low frequencies, passive coupling is not suitable
due to the size restrictions. Using double-gate transistors, this problem can be solved. The
advantages of employing these devices are:

• intrinsic separation of signal and local oscillator ports and the possibility of separate
matching;
• direct combination of the corresponding powers inside the device.

The operation mode of a double-gate MESFET can be considered equivalent to a cascade


connection [6, 7] of two single-gate MESFETs, as can be seen in Figure 5.34. The LO signal
is usually injected in the upper gate MESFET and the RF signal is applied in the lower gate
MESFET. The main operation regions are shown in Figure 5.35 where the FET1 DC curves
are shown with the FET2 DC curves inverse overlapped, but the latter have been plotted as
a function of Vg2. When a sinusoidal voltage is injected into FET2 gate, three non-linear
operating regions can be outlined (Figure 5.35):

1. Low noise mode. Defined by Vgs2 < −1 V.


2. Self-oscillating mode. Defined by −0.5 V. < Vgs2 < 1 V., Vgs1 < −1 V.
3. Image-rejection mode. Defined by 2.5 V. < Vgs2 < 3.5 V. and Vgs1 > −1.5 V.

When the transistors are biased in one of these three modes, some parts of the device act
non-linearly causing frequency conversion, while the rest acts as a RF or a IF amplifier.

1. For the first mode, the lower FET is in the linear region, while the upper FET is in the
saturation region. The mixing process takes place in the lower FET. The most important
non-linear elements are the transconductance gm and the channel resistance Rds. The upper
350 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 5.34 Cascade connection for two transistors

Figure 5.35 I/V Curves of the transistors


Mixers: Theory and Design 351

FET acts as an IF amplifier. The generated current level is quite low, and this operating
mode is therefore suitable for low noise applications.
2. For the second mode, the non-linear elements are the same as for the first mode, although
the channel resistance is a more non-linear element. The mixing takes place inside FET1,
and FET2 amplifies the IF signal.
3. For the third mode, FET1 is in the saturation region and FET2 in the linear region. The
mixing takes place in FET2, while FET1 acts as a RF preamplifier. The main nonlinearities
in FET2 are gm, Rds and Cgs.

The second mode is suitable for biasing the transistors in order to get more conversion
gain. Both transistors must be biased so that the variation of Vg2 by the LO signal changes
the gm of FET1 as much as possible. For this operation mode, the upper FET holds in
the saturation region during the LO cycle, because its Vds voltage is greater than 1.5 V.
The lower transistor is responsible for the mixing. Considering gm and Rds as the only non-
linearities of the transistor, a low frequency simplified analysis can be done. This analysis
will allow us to fix the values for both FET’s gate DC voltages (Vg1 and Vg2).
For the FET1 (in Figure 5.34) the Ids current can be expressed as:
2
 V    αVds  
3
Ids = Idss 1 − gs  (1 + λVds ) 1 −
 3  
(5.112)
 VT  
The FET1 transconductance is:

∂Ids 2
gm = =− Ids (5.113)
∂Vgs VT − Vgs

With Vgs = Vg1 the channel conductance for FET1 is:

  αV  
2

 α 1 − ds 
∂I λ  3  
gds = ds =  − Ids (5.114)
∂Vds 1 + λVds  αVds  
3

 1− 1− 
  3  

with Vds = Vds1.


FET2 holds in the saturation region during whole the LO cycle, so its current expression
can be approximated as:
2
 V 
Ids = Idss 1 − gs  (1 + λVds ) (5.115)
 VT 

with Vgs = Vg2 + VOL cos(ωpt) − Vds1. Notice that the total voltage at gate two has a DC part
(Vg2) and a AC part (VOL cos(ωpt)). As the drain currents of FET1 and FET2 are the same, we
can substitute (5.115) into the gm and gds expressions. Hence the large LO signal modulates
the transconductance and the output conductance of the FET1. Using the last Vgs expression
in Equation (5.113), we can represent gm for FET1 as a function of Vg1 and Vg2, as shown in
Figure 5.36.
352 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 5.36 Variation of gm1 (FET1) versus Vg1 and Vg2

Figure 5.37 Simplified equivalent circuit for the small signal analysis

From this figure, we can get an initial value for Vg1. The best value will allow gm to have the
largest excursion when Vg2 is varied [2]. For Vg1 values between −0.5 and 0 Volt., the gm
excursion is maximum.
If a radiofrequency small signal voltage drives the gate of the FET1 (Figure 5.37), then
the small signal current will be:

i(t) = gm(t)VRF (t) + gds(t)Vds(t) (5.116)

with

VRF = VRF cos(ω RF t)

However,

Vds(t) = −( Rds || RL)gm(t)VRF (t)

when Rds is the Rds time averaged. The IF output voltage will be:

 R AB 3R ABC 2 R 2 B3CA 
Vo (t ) =  L 2 − L 2 VOL − L 2  (VOL VRF )cos(ω FI t ) (5.117)
 VT 2V T VT 
Mixers: Theory and Design 353

where ω FI = ω p − ω s is the intermediate frequency and A, B and C are:

 −2 Idss (1 − λVds 2 ) 
A=  B = VT − Vg2 + Vdd − Vds2
 VT − Vg1 

  α (Vdd − Vds 2 )  
2

 α 1 −
RIdss λ  3  
C=  +  (5.118)
V T2 1 + λ (Vdd − Vds 2 )  α (Vdd − Vds 2 )  
3

 1− 1−
  3  
Vds2 is the drain-source voltage for the FET2, which can be considered as a constant as
this transistor remains in its saturation region throughout the LO cycle. The IF output power
will be:
1 1
Pout(ω FI) = Re(Vo(ω IF)I *(
o ω FI)) = | Vo(ω FI) |2 Re(YL ) (5.119)
2 2
Substituting (5.117) in the last expression, the intermediate frequency output power is obtained
as a function of Vg2. The best value for Vg1 has been chosen in order to maximize the gm
excursion. The variation of IF output power with external bias Vg2 is shown in Figure 5.38.
From the last figure, it is possible to choose the best value for external bias Vg2 in order to
maximize IF output power.
The expression (5.119) has been calculated considering Vds2 constant, in order to obtain a
simplified form for IF output power. Nevertheless there is a small change for Vds2 when the
external bias Vg1 and Vg2 are varied. This variation and the influence of the access resistances
have been taken into account in more elaborate analyses of the mixer.
For the FET1, the drain source current can be expressed as:
2
 V 
Ids = Idss 1 − dsi1  Tanh(αVdsi1) (5.120)
 VT1 

with Vgsi1 = Vg1 − Ids Rs, Vdsi1 = Vds1 − Ids(Rs + Rd) and VT1 = −1.32 volt.

Figure 5.38 IF output power vs Vg2


354 Microwave Devices, Circuits and Subsystems for Communications Engineering

Pout, mW
0,12

Vg1=0
0,1
Vg1=-0.25

0,08
Vg1=-0.4

0,06 Vg1=-0.5

Vg1=-0.7
0,04

Vg1=-0.8
0,02

0
-1 -0,5 0 0,5 1 1,5
Vg2, Volt.

Figure 5.39 IF output power vs Vg2

For the FET2, the drain source current can be expressed as:
2
 V 
Ids = Idss 1 − gsi 2  (1 + λVdsi 2 ) (5.121)
 VT 2 

with Vdsi2 = Vds2 − Ids(Rs + Rg), Vds2 = Vdd − Vds1, Vgsi2 = Vg2 − Ids Rs − Vds1 and VT2 = −1.8 volt.
However, the drain current for FET1 and FET2 are the same, so expressions (5.120)
and (5.121) must be identical. Solving the previous system, we can obtain the drain-source
voltage for FET1 as a function of the external bias Vg1 and Vg2. Using these values in (5.119),
we can plot the IF output power as a function of Vg2 with Vg1 as a parameter, as shown in
Figure 5.39.
A simplified model of the MESFET transistor for the F20 process of the GEC Marconi
foundry has been used to perform all the calculations. The gate width for the transistors used
was 300 µm. The parameters used are:
λ = 0.008 α = 2.52 Idss = 0.0427 A
From the simplified analysis exposed above, we can get a first design for the mixer. After
that, a final adjustment of the circuit can be done with a commercial non-linear simulator.

5.7.1 IF Amplifier
In order to get a greater conversion gain an IF amplifier has been added. A common source
amplifier has been used with a matching stage at the input. A common gate transistor has
been used as matching stage (Figure 5.40).
The transistor width has been chosen in order to match the output impedance of the mixer
to the input impedance of the amplifier, which is inversely proportional to gm. The value of
Rin can be fixed with this expression:
Mixers: Theory and Design 355

Figure 5.40 IF amplifier matching stage

−Vgs
Rin = (5.122)
Id
2
 V 
where Id corresponds to the saturation drain current: Id ≅ Idss 1 − gs 
 Vp 

5.7.2 Final Design


Using the initial values obtained in the simplified analysis, the whole mixer has been simulated
with the MDS (Microwave Design System) program, from Hewlett-Packard. The Harmonic
Balance technique has been used to simulate this mixer because it is considered more
suitable for this kind of circuit. From this simulation, load cycles have been found for both
transistors (Figure 5.41). I/V DC curves have been added in the same figure. For the upper
transistor (FET1), it can be seen that it is biased into the linear region, while FET2 is biased
into the saturation region during the whole LO cycle. Thus it is verified that the transistor

Figure 5.41 Load cycles for both transistors


356 Microwave Devices, Circuits and Subsystems for Communications Engineering

bias points are the same as the calculated ones in the simplified analysis, which allows us to
validate the design procedure.

5.7.3 Mixer Measurements


The circuit has been fabricated in the GEC Marconi foundry, which uses 0.5 µm gate length
MESFET transistors. The chip size is 2 × 2 mm2. The same design method can be used if a
hybrid implementation is desired. This circuit has been measured on wafer with a coplanar
probe station (Cascade Microtech). Figure 5.42 shows gain conversion in the IF band between
40 to 460 MHz, for an input RF signal of 1.5 GHz. Simulation results have been added in
the same figure showing a good agreement. For a 40 MHz IF signal, gain conversion has
been measured. Figure 5.43 shows measured and simulated results for a LO input power of
+7 dBm.
CONVERSION GAIN (dB)
20

Pin (LO) = +7 dBm

15

SIMULATION
10
MEASUREMENT

0
0 100 200 300 400 500 600 700
IF (MHz)

Figure 5.42 Conversion gain vs IF frequency

CONVERSION GAIN (dB)


20

15

10
MEASUREMENTS
SIMULATION
5
Pin (OL) = +7 dBm
0
400 600 800 1000 1200 1400 1600 1800 2000
RF (MHz)

Figure 5.43 Conversion gain vs RF frequency


Mixers: Theory and Design 357

OUTPUT POWER (dBm)

-50

-100
S_IM3 S_FI
M_FI M_IM3
-150

-50 -40 -30 -20 -10 0 10


INPUT POWER (dBm)

Figure 5.44 Measured and simulated IP3

LO/RF ISOLATION (dB)


35

30

25

20

15
Pin_LO=5 dBm

10
1,5 1,7 1,9 2,1
LO FREQUENCY (GHz)

Figure 5.45 LO/RF isolation vs LO frequency

Two tones test has been done to characterise the mixer intermodulation performance.
7 dBm third order interception point (IP3) of input power has been obtained. Measurement
and simulation results can be seen in Figure 5.44.
LO/RF isolation has been measured in the LO frequency band, as can be seen in
Figure 5.45. More than 20 dB has been obtained, illustrating the inherent isolation between
these ports for this topology.
358 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 5.46 180-degree singly balanced FET mixer

5.8 Single-Balanced FET Mixers


Two transistors are required to create a singly balanced mixer. Two single-device mixers
can be combined via 90° or 180° hybrids to make a singly balanced mixer. The properties of
this kind of mixers are similar to those of the singly balanced diode mixers. In order to add
the IF currents, the diodes are placed in opposite senses. However, FETs cannot be reversed,
so an IF hybrid is necessary at the output [1] (see Figure 5.46). The design of these hybrids
is the same as that of a diode mixer, and must be chosen depending on the frequency band
and the implementation type.
In the most general case, RF and LO are applied to the sum (Σ) and difference (∆) ports,
respectively, because in this case, LO signal is cancelled at the delta port. Nevertheless, RF
and LO ports can be reversed and the conversion gain and noise figure will be the same, but
the spurious-response characteristics will change. Because the IF output is derived from the
delta port, the spurious-response rejection properties of the singly balanced FET mixer are
the opposite of a singly balanced diode mixer.
Figure 5.47 shows an example of a singly balanced HEMT upconverter. The LO signal is
applied to the gates of the transistors thanks to a 180° balun, which is simply a microstrip

Figure 5.47 Singly balanced HEMT upconverter


Mixers: Theory and Design 359

T-junction with a electrical length difference of λ /2 between the two outputs. The IF input
signals are applied to the source of the transistors using a 180-degree balun. Lumped elements
were used for this balun, since the IF frequency is not high enough.
The RF signal is extracted from the drains of the transistors. Because LO signals are
applied in the opposite phase, they are cancelled at the RF output. So good LO/RF isolation
is obtained without filters. In order to eliminate the sum frequency, a band-pass filter must
be added at the output. Drains of the HEMTs are not biased, so the channel resistance of
the transistors causes the mixing. The transistor gate must be biased where the channel
resistance is most non-linear with gate-source voltage, normally near the pinch-off value. To
optimise the whole upconverter, a non-linear simulator can be used.

5.9 Double-Balanced FET Mixers


Double-balanced mixers make use of four devices for mixing. The LO and RF signals must
be applied with a 180° phase shift, so two baluns must be included to provide these phase
differences. Double-balanced FET mixers show similar properties to double-balanced diode
mixers, although the former presents conversion gain. Since there are four devices, it is not
possible to optimise or adjust the whole mixer in the same way as in single-device mixers.
On the other hand, a perfect balance between branches is very difficult to achieve in hybrid
technology, which provokes a poor performance at high frequencies.
The Gilbert cell is perhaps the best-known double-balanced topology. The basic structure
is shown in Figure 5.48. It consists of two differential pairs connected in such a way that
the different ports are mutually isolated. RF signals are injected through the gates of the
lower transistors, with a 180° phase difference. LO signals are injected through the gates of
the upper transistors, with a 180° phase difference as well. The lower transistors are biased
in the saturation zone, so they amplify the RF signal. The LO signal is injected through the
upper transistors, and this is where the mixing occurs.

Figure 5.48 Basic structure of a Gilbert cell


360 Microwave Devices, Circuits and Subsystems for Communications Engineering

14

12
T1
10

8
gm , mS

T2

0
0 1 2 3 4 5
Vg2, Volt.

Figure 5.49 Transconductance of T1 and T2 as a function of T2 gate voltage

Figure 5.49 shows the transconductance variation [7] of one of the lower transistors (T1) and
one of the upper transistors (T2) with gate voltage of T2.
The gate-source voltage of T1 transistor is chosen to place it in its saturation region.
The load line for transistor T2 (Ids vs Vds, when Vg2 is varied) has been drawn as well as the
I/V DC curves for that transistor (Figure 5.50).
M3 can be a good point since Vg2 corresponds to 1.4 volt. and it corresponds to a non-
linear zone which will give rise to a maximum conversion gain.
In order to reduce the DC consumption, the load resistors (R in Figure 5.48) were
changed to active load (transistors with gate and source short circuited). The gate width of
these transistors must be the same as the lower transistors, since the DC current is equal.
Figure 5.51 shows the final scheme of the Gilbert cell, which must be optimised using a
non-linear simulator (harmonic balance, for instance).

5.10 Harmonic Mixers


A harmonic mixer is a device where a high frequency signal (RF) is mixed with a local
oscillator signal, whose frequency is much lower than the former. An output signal is obtained
whose frequency is the difference between a harmonic of the local-oscillator and the RF
signal. Historically harmonic mixing has been used primarily at the higher millimetre wave
frequencies where reliable and stable LO sources are either not available or prohibitively
expensive. Sometimes these devices are used in frequency-multiplier design by means of phase
loop oscillators (PLOs) or as external mixers of spectrum analysers. Although theoretically
any LO harmonic can be used, second and third order are the most common, since the con-
version loss is increased with higher orders. Schottky diodes and FETs (MESFET or HEMTS)
are the mixing element most used.
Mixers: Theory and Design 361

0.1 AB
0.1 AA Vgs=0 M1
M1=42.814E–03
I1=384.81E–03
I2=
Vgs=–0.2

Vgs=–0.4

Vgs=–0.6
M1 M2 M3
Ids
Id

M4 M2
Vgs=–0.8
M2=42.531E–03
I1=685.94E–03
Vgs=–1 I2=
Vgs=–1.2
Vgs=–1.4
−0.01 A
−0.01 A

M3
M3=42.087E–03
I1=1.1538E+00
0.0E+00 vds 5.0E+00A I2=
0.0 V vds 5.0 VB

M4
M4=38.407E–03
I1=3.6240E+00
I2=

Figure 5.50 I/V DC curves and load line for the T2 transisor

Figure 5.51 Electrical scheme of the mixer


362 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 5.52 Scheme of the harmonic mixer

5.10.1 Single-Device Harmonic Mixers


Harmonic mixers using a single device can obtain the lowest conversion loss and since LO
and RF signals are very different, the design is usually quite simple. In order to obtain low
conversion loss, fundamental mixing between the signal and LO must be suppressed.
Figure 5.52 shows an example of a harmonic mixer [8] using the seventh harmonic of
the LO signal. A Schottky diode was used as the mixing element. On the RF input side,
a (λ R /4) short-circuited stub allows the signal to pass but stops the IF signal. Similarly, on
the LO and IF side the (λ R /4) open circuited stub allows the LO and IF to pass but stops
the RF signal.
A diplexer is used to inject the LO signal and to extract the IF, realised by high-pass and
low-pass filters. An inductance to ground in the low-pass filter is used as the DC return.
The RF input frequency band is 12.105–12.365 GHz and the LO signal is fixed in
1.815 GHz. The IF frequency band is 0.34–0.59 GHz. The circuit was implemented in
hybrid technology, using Cuclad 2.17 as substrate, and mounted on a metal case with
3.5 mm connectors to feed the RF, LO and IF signals. The diode used is a silicon Schottky
diode, 5082–2774, from Hewlett-Packard. The harmonic mixer was simulated using harmonic
balance and the microstrip lines were adjusted with the aid of an electromagnetic simulator
(Momentum module from Hewlett-Packard). The conversion loss is shown in Figure 5.53,
where a conversion loss of 24 dB is seen across the whole RF frequency band, with 13 dBm
LO drive. The isolation between LO and IF ports was better than 47 dB thanks to the
diplexer, and more than 50 dB of LO/RF isolation were measured in the RF frequency band.
Figure 5.54 shows a photograph of the harmonic mixer.

5.10.2 Balanced Harmonic Mixers


Although single device harmonic mixers can achieve the lowest conversion loss, balanced
topologies are commonly used, since many spurious products are rejected. Figure 5.55
shows a generic circuit of a Schottky diode-based subharmonic mixer [9]. It incorporates an
Mixers: Theory and Design 363

Figure 5.53 Conversion loss of the harmonic mixer

Figure 5.54 A harmonic mixer

Figure 5.55 Subharmonic diode mixer


364 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 5.56 Subharmonic FET mixer

anti-parallel diode pair, and the even order mixing products are suppressed (mfRF ± nfLO).
Thus the fundamental mixing product, fRF − fLO, is quenched, thereby eliminating an additional
loss mechanism and interference source.

The anti-parallel diode pair I/V curve is an odd function: I(V) = ∑C
n= 0
V 2n+1, and only
2n+1

odd order mixing products are generated at the diode pairs’ terminals.
Although the diode has been the most used non-linear element for this kind of mixer,
FETs can replace the former as shown in Figure 5.56. The LO signal is applied to the gates
with a 180° phase difference. The sources and drains are connected together. RF signal is
applied to the drains where the IF signal is also extracted, although the RF signal can be
applied to the gates if frequency bands are far enough apart.
FETs can be considered to operate in the passive or active regions. For the former, the
conductance, ∂Ids /∂Vds, is the dominant non-linearity, while the transconductance, ∂Ids / ∂Vgs,
is the main non-linearity for the latter. In both cases, the devices are operated near the pinch-
off value to achieve the highest conversion gain.
One difference between diode and FET subharmonic mixers is that the FET mixer uses
a LO balun, which is not necessary for the diode implementation. This balun represents
the major disadvantage of this circuit, especially at low frequencies. Nevertheless when
conversion gain is required, balanced FET subharmonic mixer is the best solution.

5.11 Monolithic Mixers


The gallium arsenide monolithic microwave integrated circuit (GaAs MMIC) technology
has undergone great development in the past twenty years, achieving a maturity grade
similar to that of the silicon technology [10]. A monolithic circuit is one where all compo-
nents, both passive and active, are incorporated into a single semi-conductor die allowing
complete operation by the application of DC and microwave signals. Thus, very little wire-
bonding and assembly is required and the size and weight of the circuits are usually much
smaller, allowing each subsystem size reductions within the same volume as occupied by a
hybrid circuit. Monolithic circuits can be fabricated in large quantity at low cost, allowing
their use in consumer applications. Although the prices have been decreasing in the last
years, the major limitation is the high cost of prototypes. Only for large quantities, are
the prices comparable to the hybrid circuits. Perhaps the major advantage to be gained
Mixers: Theory and Design 365

from monolithic microwave technology will come about when high levels of integration
can be produced at affordable costs with acceptable performance. Many circuit functions
now available with MMICs would have been impossible to produce using conventional
substrate-based hybrid technology. This is particularly true in the case of circuits requiring
many different gate-width FETs. Although the hybrid circuits are still being used in many
applications, MMICs have begun to make up an important part of many available micro-
wave products.

5.11.1 Characteristics of the Monolithic Medium

Due to the great development in monolithic technology MMICs are being used in many
applications. The main reasons for improved performance of MMICs over their hybrid
counterparts include the following:

• The assembly interconnects are eliminated, which reduces the parasitics due to bond
wires.
• The reliability of the circuits can be improved owing to the much-reduced number of
interconnections.
• All the components can be optimised to meet the needs of the circuit performance,
without being limited to a discrete catalogue of components.
• The cost of each MMIC does not depend on the number of active elements, as in hybrid
technology.
• Highly reproducible performance that provided by batch processing is possible and
• All circuit functions can be integrated.

Nevertheless, there are some disadvantages compared to hybrid circuits. Considerable invest-
ment is required in manufacturing facilities and staff in order for a company to produce
its own circuits. Although the technology has improved, MMIC prices are only of interest
for circuits that will be made in large numbers. On the other hand, the fabrication of the
monolithic circuit is the end of its design cycle. No adjustment can be made after the circuit
is finished. This fact requires us to model the components of the circuit in a very accurate
way. In fact, the designer is limited by the precision of these models, above all by those
of the active elements. The modelling of active components is not very different from
modelling hybrid circuits. As there is great freedom in adjusting the geometries of FET
devices, the modelling of such components can be more difficult.
For mixers, the non-linear performance of the devices that produce the mixing must
be characterised in an accurate way, in order to predict the conversion losses of the circuit.
This is a difficult task for some devices, so the performance prediction in mixer circuits is
more complicated than for other circuits. Passive elements are sometimes measured and
their s-parameters stored in a database, which are then used in the design.
From all the chips produced on the same wafer, only a proportion of them will operable
due to imperfections in processing. The term yield refers to the number of circuits on a
given wafer that deliver acceptable electrical performance. Clearly the cost of a chip is
inversely proportional to yield, and thus designers must avoid structures that cannot be
produced reliably.
366 Microwave Devices, Circuits and Subsystems for Communications Engineering

5.11.2 Devices

MMIC mixers can be classified into transistor and diode structures. If GaAs MESFET techno-
logy is considered, the diode mixer is made with a FET, using the junction between the gate
and the channel. Although the cut-off frequency is higher for the diodes, the width and the
geometry of the MESFETs can be modified in order to obtain the desired specifications. Its
is nearly impossible to find this degree of freedom with discrete devices in hybrid circuits.
The MESFET transistor is commonly used in monolithic technology for frequencies above
1 GHz. Active MESFET mixers offer many advantages over passive ones. This is especially
true for double-gate MESFETs, because there is an inherent isolation between the RF and
LO ports.
However, there are several drawbacks when designing MESFET mixers. If a diode is used
as the non-linear element, it is possible to obtain a good first-order approximation with
a linear analysis. But, with the MESFET mixers, the analysis becomes very complicated.
Moreover, when there are several transistors, it is necessary to use computer-aided design in
order to predict the performance of the mixer.
There are some applications where mixers can be done with FETs but not with diodes.
Sometimes, this advantage is due to the compatibility of the FET transistor with the monolithic
circuit processes. However, in other cases, it is due to the inherent advantages of the FET
over the diode mixers.
More recently HEMTs and HBTs have appeared; the former uses a GaAs’ substrate
and the latter can use either GaAs or silicon. Both of them have a common characteristic:
a heterojunction is used in its construction. These heterojunctions are formed between
semiconductors of different compositions and band gaps. The properties of these new
devices are superior to those of the MESFET in having a higher cut-off frequency, greater
gain, and lower noise figure. As these processes are newer, one might suppose that yield
will be lower than in MESFET processes. Much effort has been put in over recent years
in order to improve yield of these processes. Nowadays, although the maturity of these
technologies is not that of the silicon ICs, we can hope that it will be possible in the
near future.

5.11.3 Single-Device FET Mixers

The design procedure used for these mixers is the same as that of the hybrid mixers.
Single-device mixers usually have greater conversion gain than balanced topologies and the
optimisation method is easier. In order to avoid using a large substrate area, lumped-element
circuits can be used when the frequency band is not very high.
Figure 5.57 shows an example of a single-device HEMT mixer [11]. This topology
corresponds to a drain mixer, since the LO signal is injected by the drain. RF and LO
signals were applied to separate terminals because their frequency bands were very close.
As explained in Section 5.6.1.3, when the LO signal is injected through the drain, the
transconductance gm and the channel resistance Rds are modulated and become time-varying
functions, with a period equal to that of the LO signal. The maximum conversion gain is
given by (see Section 5.6.1.3 for more details):
Mixers: Theory and Design 367

| µ1 | 2
Gm = (5.123)
4ω 12 Cgs2 ( Rgm + Rs )( Rdr + Rs + Rdo )

Where µ = gm Rds is the voltage amplification factor.


Therefore, the fundamental component of this voltage amplification factor must be max-
imised in order to get as great a conversion gain as possible. For this purpose it is necessary
to find a bias point for the transistor in which µ is more sensitive to the bias modulation by
the LO signal. For a drain mixer, the channel resistance non-linearity has the same contribu-
tion to the mixing as the tranconductance non-linearity. For this reason, both of them must
be taken into account. Figure 5.58 shows Gm versus drain-source voltage with gate-source
voltage as a parameter (following Section 5.56).
At the RF input a lumped-elements low-pass filter structure has been used. This structure
has similar properties to those of the ladder transformers, but it can be built in a compact
way, even at low frequencies. The main difference between this topology and a low pass
filter is that the resistances at the terminals can be different, which means that the reflection
losses at zero frequency will be reasonable. In this case, lumped elements were used to
implement the input and output networks since the working frequency is not very high and
these networks would occupy a large substrate area if distributed elements were used. To
bias the transistor gate, a lumped-element circuit was designed, also creating a short circuit
at IF on the RF port. This circuit down-converts 14–17 GHz RF signals to a 1 GHz IF band
with 2.5 dB of conversion gain by using 15 dBm of LO drive. Table 5.1 summarises the
measured results of this circuit.
For the LO and the IF signals a diplexer has been designed, since both frequency bands
are very distant in frequency and good isolations can be achieved. Lumped elements were
used for the same reason than in the RF input network.
This mixer was fabricated by Philips Microwave Limeil (France) using the D02AH pro-
cess, which uses 0.2 µm of gate length in the HEMT transistors. The chip size is 1.5 mm2,
and a photograph of the mixer is shown in Figure 5.57.

Table 5.1 14–17 GHz MMIC single-ended mixer

RF bandwidth 14–17 GHz

RF bandwidth 1 GHz
NF (SSB) 7.6 dB (15 GHz)
Gain 2.5 dB
Isolation LO to RF > 23 dB
LO to IF > 42 dB
Return Losses (LO, RF) > 10 dB
LO power 15 dBm
368 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 5.57 The mixer

5.11.4 Single-Balanced FET Mixers


This kind of mixer is widely used in monolithic circuits. Thanks to the homogeneous
performance of the components fabricated in the same wafer, the isolations between ports is
higher than in hybrid technology. In the latter case balanced mixers often use large passive
structures, such as Branch-Line or Rat-Race couplers, to form the baluns. Nevertheless, a
large substrate area is required if monolithic implementation is desired. Lumped elements
are another option to make these distributed baluns, but they generally present low band-
width and higher losses.
Figure 5.58 shows an example of singly balanced mixer [12]. Mixing and balun functions
are carried out by the same structure. This topology is similar to that of the centre tapped
balun. Two transistors in gate-common, source-common structures realise the mixing and
the phase shifting, thanks to the way they are connected.

Figure 5.58 Singly balanced FET mixer


Mixers: Theory and Design 369

For the common gate HEMT, the LO signal is injected between the gate and source ports.
In this case, the time-varying transconductance is the dominant contributor to frequency con-
version, and the effect of other non-linearities is minimal. The conversion gain is proportional
to the fundamental LO-frequency component of such a transconductance waveform gm(t). In
a HEMT transistor, gm is the maximum when Vds is the maximum, so its drain voltage must
remain in the saturation region throughout the LO cycle.
In order to avoid a drop in the drain/source resistance and an increase in gate/source
capacitance, it is important to guarantee that Vds does not decrease below the knee voltage.
Otherwise, the conversion gain will decrease and the noise figure will increase. For the
common-gate transistor, the input impedance is proportional to 1/gm. The width of this
transistor is chosen to match the combiner output impedance to the common-gate transistor
input impedance.
For the common-source transistor, the time-varying transconductance is also the dominant
contributor to frequency conversion. The maximum conversion gain is [4]:

g12 Rd
Gc = (5.124)
4ω 12 A 2 Rin

Where Rin = Rg + Ri + Rs, A is the average value of Cgs and Rd is the average value of Rds.
g1 is the magnitude of the fundamental component of the transconductance waveform. In
order to maximise g1, the transistor must be biased near the pinch-off value. The variation
of gm with the gate voltage for a 300 µm gate width is shown in Figure 5.59. In the same
figure the variation of g1 as a function of the gate voltage was added. It can be seen that
maximum value of g1 corresponds to the device turn-on voltage. From both figures, it can
be seen that the g1 maximum (13 mS) is 1/3.3 of gm maximum (43 mS). This ratio is very
close to the 1/B ratio obtained for the ideal case, when the gm is a step function of gate
voltage.
By selecting the gate width of the common-source transistor, it is possible to have the
same gain in both devices (common-source and common-gate), and IF and LO signals will
be cancelled at the output of both transistors due to the phase shift (ideally 180°) in the
common-source transistor. This is true in low frequencies but for higher frequencies, the

(A) (B)
50 15

40 12
Gm (mS)

g1 (mS)

30 9

20 6

10 3

0 0
-1,5 -1 -0,5 0 0,5 -1,5 -1 -0,5 0
Vgs (volt.) Vgs (Volt.)

Figure 5.59 (a) gm vs Vgs. (b) g1 vs Vgs


370 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 5.60 The singly balanced FET mixer

phase shift begins to differ from 180°, decreasing the isolation between output and input
ports. When LO and RF signals are very close in frequency, it will be easier to optimise the
mixer at both frequency bands. Nevertheless, if LO and RF frequencies are very different,
that is the case of an upconverter, it will not be possible to have good balance (phase and
amplitude) at both frequencies, and filters will be necessary in order to achieve good
isolations.
The combiner for LO and RF signals is formed by two amplifiers, providing good isolation
between both ports, and furthermore, amplifying both signals. Lumped elements were used
to match amplifier input circuits.
The output network can be different depending on whether the mixer is to be a down-
converter or an upconverter. For the first option, a matching network is sufficient, but for an
upconverter a high-pass filter will be necessary for the reasons explained above.
The mixer of the example up-converts 1.885 GHz IF signal to a 14–14.25 GHz RF band
with 4.2 dB of conversion gain by using only 3 dBm LO drive. This mixer was fabricated
in Philips Microwave Limeil (France) using the D02AH process, which incorporates 0.2 µm
of gate length for the HEMT transistors. Figure 5.60 shows a photograph of the mixer.
The chip size is 3 mm2.

5.11.5 Double-Balanced FET Mixers

When the input and output frequency bands are very close, or for very broadband applica-
tions, double-balanced mixers can obtain the best performances. This is especially true in a
monolithic implementation, where the dispersion is equal for similar elements. This means
that all the transistors in the chip will have the same performance, and the same can be
said for the capacitors, inductors, etc. on the chip. This property produces double-balanced
Mixers: Theory and Design 371

Figure 5.61 Active balun

mixers with better isolations between the ports when they are fabricated in monolithic
technology.
The design problem for balanced mixers can be divided into two main areas: the non-
linear element and the balun. If a monolithic implementation is desired, the balun dimension
is limited by the chip area, especially for low frequencies (below 20 GHz). Thus, active
balun or lumped-element transformers are the only viable options. An active balun is shown
in Figure 5.61. This circuit uses the known property of a transistor amplifier, where the
phase shift between drain and source ports is 180° ideally. This property is only true at low
frequencies. When the frequency begins to increase, the phase shift change due to capacitance
and inductance affects the transistors.
In a high frequency analysis, and using a unilateral electrical model for the MESFET
transistor, we can get:

V1 − gm Rds − ω 2 RRdsCgsCds + jω RCgs


= (5.125)
V2 gm Rds − ω 2 RRdsCgs + jω ( Rds + RCgs )

where:

Rds: channel resistance


gm: transconductance,
R: the load resistance at the output
Cgs: the gate-source capacitor
Cds: the drain-source capacitor.

There is other balun topology which can be implemented using monolithic technology, and
can eliminate these problems. Two source common amplifiers coupled by their sources to
a resistance or current source are used (Figure 5.62). This topology, commonly used in
low frequency circuits with bipolar transistors, can be translated to microwaves frequencies
using MESFET transistors as amplifier element. In the traditional differential stages, the
basic function is to amplify the difference between two input signals. Nevertheless, one of
the inputs is connected to ground using a capacitor if the circuit must operate as a balun.
The outputs are taken from the transistor drains.
Although this topology obtains better results, when the frequency increases, the phase
errors increase as well, which makes the circuit not very useful for microwave balanced
mixers. Better results can be obtained if a second stage is connected at the output.
372 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 5.62 Balun topology

By using this last topology as balun, a double-balanced mixer was designed [7, 13]. A
MESFET ring was used as mixer and two differential pairs as RF and LO baluns. In order
to get low intermodulation levels, cold FETs were used as the mixing element. The channel
resistance of a MESFET transistor without biasing can vary when a signal is applied to the
gate. For a non-biasing transistor, the drain current is nonlinear with Vgs, which allows
the channel resistance to change with time. Nevertheless, the drain current is almost linear
with Vds. Thanks to this last characteristic the distortion level generated is very low when
compared to that generated by a diode. Thus, the LO signal is applied to the gate and the
RF signal to the drain.
The IF signal is filtered from the drain for single-ended mixers, but for balanced topologies
it is obtained from the source. In both structures (single or balanced), there is a problem if
the transistor is biased at zero volts. The capacitance between the drain and gate terminals
increases until values close to the Cgs capacitance, which would provoke some coupling
between LO and RF ports and diminish the isolation between these terminals. Also, if LO
signal is coupled to drain port the voltage at this terminal would greater than zero, increasing
the intermodulation level. It is possible to avoid these problems by using filters in both ports.
However, for broadband applications balanced topologies must be used, as it is shown in the
example of the Figure 5.63.
The variation of channel conductance with gate-source voltage, for a 400 µm gate width
MESFET, is shown in Figure 5.64. The mixer transistors are biased in the most non-linear
region in order to get the lowest conversion losses. As the channel conductance is being
modulated by the LO signal, Gds can be developed in a Fourier series:

Gds = g0 + 2g1 cos(ωLO t) + . . . (5.126)

where ωLO is the local oscillator frequency.


g1 must be maximised in order to get the minimum conversion losses. This means that the
transistors will be biased near the pinch-off voltage (≅ −1.7 volt.). This gate biasing point is
Mixers: Theory and Design 373

Figure 5.63 Double-balanced FET mixer

Figure 5.64 Gds versus Vgs

the most convenient, since small variations of Vgs provoke large changes of Gds. It means that
this is the most sensitive biasing point for the channel resistance with respect to gate voltage
changes.
In order to get a low distortion level, a RF balun should be designed with a view to not
degrading the IM performance of the mixer.
The double-balanced mixer of the example down-converts 1.8 GHz of RF frequency
to 40–860 MHz IF band with 2 dB of conversion losses. LO/IF isolation is shown in
Figure 5.65 as a function of the transistor gate voltage. As we can see in the figure, more than
40 dB is achieved, showing the good isolation between ports in double-balanced topologies.
374 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 5.65 LO/IF isolation vs Vgs

Figure 5.66 Performance of the double-balanced mixer

Low level intermodulation has been measured, as can be seen in Figure 5.66, in the two-tone
test, thanks to the cold FET mixer. This downconverter was fabricated by Philips Micro-
wave Limeil (France), using the ER07AD process, which incorporates 0.7 µm of gate length
for the MESFET transistors. Figure 5.67 shows a photograph of the whole downconverter,
where the chip size is 3 mm2.
Mixers: Theory and Design 375

Figure 5.67 The double-balanced mixer

Appendix I

1 Y k! 
cosk x = ⋅ ∑ ⋅ cos((k − 2 y) ⋅ x ) + bk 
2 k−1  y = 0 (k − y)! ⋅ y! 
where

k −2
 for k even and k≠0 1 k!
2  2 ⋅ (k / 2!)2 for k even and k≠0
Y = and bk = 
k −1 0
for k odd  for k odd
 2

k is a natural number and k − 2y is always greater than zero.

Appendix II: Modified Bessel functions of first species and n order


n +2 ⋅ k
 x

 2
In ( x ) = ∑
k =0 k! ⋅ Γ(n + k + 1)
Where Γ(n + k + 1) = (n + k) · Γ(n + k) = (n + k)!

and
x2 x4 x6
I0 ( x ) = 1 + + 2 2 + 2 2 2 +...
2 2
2 ⋅4 2 ⋅4 ⋅6
x2 x4 x6
I1( x ) = 1 + + 2 2 + 2 2 2 +...
2 2
2 ⋅4 2 ⋅4 ⋅6
376 Microwave Devices, Circuits and Subsystems for Communications Engineering

I0(x) Function

x 0 1 2 3 4 5 6 7 8 9

0. 1.000 1.003 1.010 1.023 1.040 1.063 1.092 1.126 1.167 1.213
1. 1.226 1.326 1.394 1.469 1.553 1.647 1.750 1.864 1.990 2.128
2. 2.280 2.446 2.629 2.830 3.049 3.290 3.553 3.842 4.157 4.503
3. 4.881 5.294 5.747 6.243 6.785 7.378 8.028 8.739 9.517 10.37
4. 11.30 12.32 13.44 14.67 16.01 17.48 19.09 20.86 22.79 24.91
5. 27.24 29.79 32.58 35.65 39.01 42.69 46.74 51.17 56.04 61.38
6. 67.23 73.66 80.72 88.46 96.96 106.3 116.5 127.8 140.1 153.7
7. 168.6 185.0 202.9 222.7 244.3 268.2 294.3 323.1 354.7 389.4
8. 427.6 469.5 515.6 566.3 621.9 683.2 750.5 824.4 905.8 995.2
9. 1094 1202 1321 1451 1595 1753 1927 2119 2329 2561

I1(x) Function

x 0 1 2 3 4 5 6 7 8 9

0. 0.000 0.050 0.100 0.151 0.204 0.257 0.313 0.372 0.433 0.497
1. 0.565 0.637 0.715 0.797 0.886 0.982 1.085 1.196 1.317 1.448
2. 1.591 1.745 1.914 2.098 2.298 2.517 2.755 3.016 3.301 3.613
3. 3.953 4.326 4.734 5.181 5.670 6.206 6.793 7.436 8.140 8.913
4. 9.759 10.69 11.71 12.82 14.05 15.30 16.86 18.48 20.25 22.20
5. 24.34 26.68 29.25 32.08 35.18 38.59 42.33 46.44 50.95 55.90
6. 61.34 67.32 73.89 81.10 89.03 97.74 107.3 117.8 129.4 142.1
7. 156.0 171.4 188.3 206.8 227.2 249.6 274.2 301.3 331.1 363.9
8. 399.9 439.5 483.0 531.0 583.7 641.6 705.4 775.5 852.7 937.5
9. 1031 1134 1247 1371 1508 1658 1824 2006 2207 2428

References
[1] S.A. Maas, Microwave Mixers, 2nd edn, Artech House, MA, 1993.
[2] S.A. Maas, Non Linear Microwave Circuits, Artech House, MA, 1989.
[3] R.A. Pucel, D. Massé, R. Bera. ‘Performance of GaAs MESFET Mixers at X Band’, IEEE MTT, vol. MTT-
24, no. 6, June 1976, pp. 351–360.
[4] G.D. Vendelin, D.M. Pavio, V.L. Rohde, Microwave Circuit Design. Artech House, MA, 1979.
[5] S. Balatchev, J.L. Gautier, B. Delacressonnière, ‘Using a negative conductance for optimizing the resistive
mixers’ conversion losses’, Galium Arsenide Applications Symposium, GAAS 96, 4C5.
[6] C. Tsironis, R. Meier, R. Stahlamann, ‘Dual-Gate MESFET Mixers’, IEEE MTT, vol. MTT-32, no. 3, March
1984, pp. 248–255.
[7] M.L. de la Fuente, ‘Diseño de mezcladores de microondas en tecnología monolítica’, PhD thesis, Universidad
de Cantabria, November 1997.
[8] F. Diaz, A. Herrera, E. Artal, A. Tazón, M.L. de la Fuente, J.M. Zamanillo, F. López, ‘Diseño de un PLO
Sintetizado en Banda Ku’, URSI XI Simp. Nac., Madrid, Sept. 1996.
Mixers: Theory and Design 377

[9] A. Madjar, ‘A novel general approach for the optimum design of microwave and millimeter wave subharmonic
mixers’, IEEE MTT, vol. 44, no. 11, November 1996, pp. 1997–2000.
[10] R. Goyal, Monolithic Microwave Integrated Circuits, Artech House, MA, 1989.
[11] M.L. de la Fuente, J. Portilla, E. Artal, ‘Low noise Ku-band drain mixer using P-HEMT technology’, paper
presented at IEEE, 5th International Conference on Electronics, Circuits and Systems, Lisbon, September
1998, pp. 175–178.
[12] M.L. de la Fuente, J. Portilla, J.P. Pascual, E. Artal, ‘Low-noise Ku-band MMIC balanced P-HEMT upconverter’,
IEEE Solid-State Circuits, vol. 34 February 1999, pp. 259–263.
[13] M.L. de la Fuente, J.P. Pascual, E. Artal, ‘Low intermodulation converter system for TV distribution’, paper
presented at XII Design of Circuits and Integrated Systems Conference, Seville, Spain, November 1997.
378 Microwave Devices, Circuits and Subsystems for Communications Engineering
Filters 379

6
Filters
A. Mediavilla

6.1 Introduction
Filter circuits are a key component in any high frequency wireless system. Modern trends
have been to try to move away from analogue filtering as much as possible and to implement
filtering using digital signal processing wherever possible. However, this is only achievable
at lower frequencies where the analogue signal can be successfully transformed into the
digital domain. In processing signals at microwave frequencies, the standard approaches to
filter designs will still be required for the foreseeable future.
This chapter describes the different types of filter (low pass, high pass, band pass and band
stop) and their characteristic responses (Butterworth, Chebyshev, Bessel and Elliptic). To
enable a generalised approach to filter design, given any particular specification, the chapter
describes how a low pass prototype filter can be transformed into any of the other types of
filter and how the frequency response can also be scaled to fit the defined corner frequencies.

6.2 Filter Fundamentals


6.2.1 Two-Port Network Definitions
Let us consider a general microwave two-port network, with generator and load termination
RG and RL respectively, under sinusoidal operation, as shown in Figure 6.1.
In this case, and assuming that the source generator is Eg, we can define the power content
in the network as a function of the terminal voltages and currents as follows:

| Eg | 2
Pin = Pdiss = 1/2 Re{V1 · I*1 } Pref = Pin − Pdiss
8 RG
(6.1)
| V2 | 2
PL = 1/2 Re{V2 · (−I*2 )} = 1/2 = 1/2 RL · | I2 |2
RL

where Pin, Pdiss, Pref and PL are the incident or available, dissipated at the input, reflected, and
output power. All the voltages and currents in the equations are in phasor form (complex

Microwave Devices, Circuits and Subsystems for Communications Engineering Edited by I. A. Glover, S. R. Pennock
and P. R. Shepherd
© 2005 John Wiley & Sons, Ltd.
380 Microwave Devices, Circuits and Subsystems for Communications Engineering

RG I1 I2

Eg
(a) +
V1
Two V2 RL

Port
RG

(b)
Eg
+
Pin
Pdiss
Two Pout RL

Pref Port
Zin Γin

Figure 6.1 Two-port microwave network definitions

values where the modulus is the peak value), and the input impedance of this network is
defined as:
Zin = V1 /I1 (6.2)
In the same way we can define the concepts of voltage gain, power transfer gain and
attenuation:
Voltage gain: Av = V2 /V1 (6.3)
Av | dB = 20 · Log | Av | (6.4)

Power transfer gain: GT = PL /Pin (6.5)


GT | dB = 10 · Log(GT) (6.6)

Network attenuation: Atn = Pin /PL = 1/GT (6.7)


Atn | dB = 10 · Log(Atn)
Furthermore, we can introduce the high frequency concepts such as reflection coefficient
at the input, return loss and their relationship with the power content and the scattering
parameters of the network.
The reflection coefficient Γin at the input (referred to RG) is defined as:
Zin − RG
Γin = (6.8)
Zin + RG
where its modulus ρ is the ratio between the incident power and the reflected power at the
input of the network:

Zin − RG
ρ = | Γin | = ρ 2 = Pref /Pin (6.9)
Zin + RG
Filters 381

Return loss:

Rloss = Pref /Pin = ρ 2 = 1 − Pdiss /Pin (6.10)

Scattering parameter S11:

S11 = Γin | S11 | 2 = ρ 2 = Rloss (6.11)

Scattering parameter S21:

| S21 | 2 = PL /Pin = GT (6.12)

If the two-port is a lossless network, the dissipated input power Pdiss and the output power
PL are the same because inside the network there is no dissipating device such as resistors.
In this case, we can write the following modifications:

1
GT = 1 − ρ 2 Atn = (6.13)
1 − ρ2

Rloss = 1 − GT = 1 − 1/Atn (6.14)

6.2.2 Filter Description


Generally speaking, a filter is any passive or active network with a predetermined frequency
response in terms of amplitude and phase.They can be classified, depending on their applica-
tion, as low pass, high pass, band pass and band stop filters, as shown in Figure 6.2.
Low pass and high pass filters are characterised by their cut-off frequency Fc where the
transfer gain usually drops to one half (3 dB point). Conversely, band pass and band stop
filters are defined by their centre frequency F0 along with their 3dB bandwidth BW3.
It seems obvious that in a filter design it is not enough to define the cut-off frequencies or
3dB bandwidths. Most applications require a given attenuation at a given frequency – out of

Low Pass Band Pass


G T (dB) G T (dB)
LPF 0 BPF 0
–3 –3

BW3 BW3

F F
Fc F0
High Pass Band Stop
G T (dB) G T (dB)
HPF 0 BSF 0
–3 –3

BW3 BW3

F F
Fc F0

Figure 6.2 Types of filter


382 Microwave Devices, Circuits and Subsystems for Communications Engineering

G T (dB)
Pass band Stop band
0
–3

BW3

5 N=3
4
–40

F(GHz)
2.4 2.56

Figure 6.3 Variation of stopband characteristics with filter order number

band – while maintaining the cut-off values and the 3dB bandwidth. This new requirement
will condition the filter response and the number of sections (complexity).
In the Figure 6.3 we have several low pass filters having the same cut-off frequency of
2.4 GHz with a different number of sections, N.
As we increase the number of sections N, we have a steeper transition from the passband
to the stopband while maintaining the same passband characteristics and cut-off frequency.
If our specification at, for example, 2.56 GHz is that the attenuation should be better than
40 dB, it is necessary to use an order N > 4.
Apart from the order of a filter, we have another important characteristic, that is the
filter response: there are many types of frequency responses, and the choice depends on the
application. The most important are:

• Butterworth: maximally flat in amplitude;


• Chebyshev: amplitude passband equi-ripple;
• Bessel: maximally flat in phase;
• Elliptic: amplitude passband and stopband equi-ripple.

The choice of one of the above characteristics depends mainly on the system application
where the filter should be used as well as the type of signal that is passing through the filter:

• pure sinusoidal;
• square (train of pulses);
• AM or FM modulation;
• more complex modulations.

In Figure 6.4 we have three different band pass N = 3 filters at the same centre frequency
F0 = 900 MHz and with the same 3dB bandwidth BW3 = 60 MHz.
The Elliptic response exhibits a more abrupt transition but the stopband has a residual
ripple. The Bessel response has a very poor amplitude response but it has an excellent phase
behaviour. Finally, Butterworth and Chebyshev filters can be considered as a compromise
between the other two. In the passband Chebyshev and Elliptic filters have an equi-ripple
response while Bessel and Butterworth filters have a flat behaviour.
Filters 383

G T (dB) N=3 filters


Stop band Pass band Stop band
0
–3
Chebyshev
Butterworth
Elliptic
BW3 Bessel
60 MHz

F(MHz)
900

Figure 6.4 Different filter response types

6.2.3 Filter Implementation


All the filters described above, having their own particular characteristics, can be built in
very different ways and technologies. The choice mainly depends on the working frequency,
filter selectivity and losses in the passband.
In the low frequency range (up to several MHz) most filters are designed on the basis of
operational amplifiers: active filters. This is due to the fact that an implementation with LC
networks leads to LC values out of commercial range, apart from the design flexibility when
using Opamps.
In the radiofrequency band (up to a few GHz) filter synthesis is accomplished by using
classical discrete RLC elements, or more recently high Q coaxial resonators. The only
inconvenience here are the lumped-element losses when increasing the frequency.
When we want to design up to 30/40 GHz, filter synthesis does not use lumped elements
but coaxial, planar transmission line technologies (microstrip, strip, slab, coplanar lines)
or waveguide technology (rectangular, circular, ridge, etc.). The discrete L and C elements
can be built by using sections of transmission lines when the bandwidth is not too low, or
alternatively dielectric resonators for high Q filters.
Finally, at frequencies above 40 GHz (high frequency radio links, radio astronomy, etc.),
the losses in coaxial and planar technologies are so high that it is necessary to use waveguide
configurations.

6.2.4 The Low Pass Prototype Filter


It is obvious that the mathematical techniques for filter synthesis cannot look for the infinite
variety of cut-off frequencies, centre frequencies, 3dB bandwidths that we can imagine, as
well as whether the filter is a high pass or stop band, etc.
Consequently, all the filter synthesis techniques and filter shapes refer to the prototype
low pass filter (PLPF). This normalised filter shown in Figure 6.5 has a cut-off frequency
of 1 rad/sec (0.159 Hz) for a general 3dB attenuation.
We will see in the next section that we can synthesise PLPF with Butterworth, Bessel,
Chebyshev, etc. characteristics. We also see how we can scale in frequency this PLPF in
384 Microwave Devices, Circuits and Subsystems for Communications Engineering

G T (dB) Atn(dB)
Pass band Stop band Pass band Stop band
0
–3

BW3

BW3

␻ 3 ␻
0
1rad/s 1rad/s

Figure 6.5 Normalised LPF characteristics

order to have LPF at the frequencies of interest, and how to go directly from a PLPF to
an HPF, BPF or BSF at a given frequency. Finally, we will see how the actual out of band
specifications can be translated into PLPF constraints. In conclusion, our problem is reduced
to mathematically synthesising prototype filters PLPF with specified attenuation in the
passband and stopband.

6.2.5 The Filter Design Process


Starting from the above considerations, the methods used in filter design can be summarised
as follows:

1. The designer has several filter specifications (inband and outband) as well as system and
incoming signal constraints. The first step is to select one or two types of filter that could
meet the specifications along with the presumed technology.
2. The frequency response specifications are transformed into PLPF specifications with a
cut-off frequency of ωc = 1 rad/s.
3. We compare in the PLPF plane the desired response with the theoretical responses
(Bessel, Chebyshev, etc.) and we select the minimum number of sections and the type of
filter.
4. We perform frequency scaling and transformations in order to arrive at the specified
frequency range.
5. We implement the filter into the specified technology, depending on the frequency range.

6.2.5.1 Filter simulation


The most important measures that a filter designer should have at his or her disposal are
obviously the attenuation (or the inverse transfer gain), the return loss in order to see the
frequency matching, and more specifically the group delay if the input signal has any kind of
modulation. We must remember that the attenuation is directly related to the S21 parameter
while the return loss is related to the S11.
Although many commercially available simulators have implemented all the above
measures, other popular simulators such as PSPICE do not have these facilities: they do not
have the concept of scattering parameters. Therefore, it is of interest to arrange the circuit
implementation in order to directly measure the key parameters.
Filters 385

1 RG 2 4 5 = S21

+
– Subcircuit
VG2
+ AC 1V + 1/2V(4).
1
– 3 = S11
XNET RL – (RG /R)
V G1
AC 2V
1012

Figure 6.6 PSPICE configuration for generating s-parameters

Figure 6.6 shows a PSPICE configuration able to generate the scattering parameters of
a two-port network measured with respect to any RG and RL (resistive source and load
impedances).
The AC voltage at node 3 is exactly (in modulus and phase) the S11 parameter while the
same applies for the node 5 and the S21 parameter. If we want to calculate S22 and S12, we
have to turn the two-port network. Figure 6.7 shows the details of the circuit description in
PSPICE along with the measurements to be performed.

6.3 Mathematical Filter Responses


As has been stated above, our interest now is to design and to control prototype low pass
filters PLPF with different shapes of pass band and stop band attenuation. In this section
we will study the commonly used characteristics: Butterworth, Chebyshev, Bessel and
Elliptic, although many other characteristics are available in the literature for special
applications.

6.3.1 The Butterworth Response

Commonly known as a ‘maximally flat’ filter, the attenuation has a frequency response
where the N − 1 first derivatives at DC (ω = 0) are null, N being the order of the filter. This
can be written as shown is Equation (6.15):

Atn = 1 + (ω)2N Atn | dB = 10 · Log[1 + (ω)2N ] N = order (6.15)

From this equation we can immediately see that:

at ω = 0 Atn = 1 Atn | dB = 0 (independent of the order N)


at ω = 1 Atn = 2 Atn | dB = 3 (independent of the order N)
386 Microwave Devices, Circuits and Subsystems for Communications Engineering

PSPICE Circuit Description

.PARAM RG=100 RL=50

VG1 1 0 AC 2V
VG2 2 3 AC 1V
RAUX 3 0 1.E12
RG 1 2 {RG}
RL 4 0 {RL}
EAUX 5 0 VALUE={ V(4) * SQRT(RG/RL) }
REAUX 5 0 1.0
XNET 2 4 Name of Subcircuit

.AC Frequency Sweep

PSPICE Measurements with PROBE

VM(3) ------------------------ Amplitude of S11


VP(3) ------------------------ Phase of S11
VM(5) ----------------------- Amplitude of S21
VP(5) ------------------------ Phase of S21
VDB(3) ---------------------- Return Loss (dB)
VDB(5) ---------------------- Transfer Gain (dB)
-VDB(5) --------------------- Attenuation (dB)
-1/360 * d(VP(5)) -------- Group delay (sec)

Figure 6.7 PSPICE circuit description for s-parameter generation

at ω Ⰷ 1 Atn grows towards infinity at a ratio of 6N dB/octave because the nominal value
grows as ω 2N.
at ω = 0 d n(Atn)/dω n = 0 for 0 < n < N (maximally flat condition)
Figure 6.8 shows the typical response of these filters as a function of the number of sections N.
We can observe that the cut-off frequency is always ωc = 1 rad/s for an attenuation of
3 dB, independent of N. As the order N increases, the attenuation in the stopband area
grows while the flatness in the pass band area is more evident. This kind of filter is an ‘all
pole filter’ because infinite attenuation is reached with ω going to infinity.

6.3.2 The Chebyshev Response


This characteristic has, for the same order N, a higher degree of stop band attenuation
than the Butterworth response. This can be accomplished by allowing a certain amount
of controlled ripple in the pass band. That is why these kinds of filters are called ‘ripple-
controlled filters’ and, as well as the Butterworth filters, are ‘all pole filters’. The attenuation
can be written as shown in Equation (6.16):
Atn = 1 + K 2 · T N2 (ω) Atn | dB = 10 · Log[1 + K 2 · TN2(ω)] N = order (6.16)
Filters 387

Atn(dB)

4
3

N=2

BW3

ω
0
1 rad/s

Figure 6.8 Butterworth LPF responses as a function of N

where TN is the Chebyshev polynomial of degree N:


T1(X) = X T5(X) = 16X 5 − 20X 3 + 5X
T2(X) = 2X 2 − 1 T6(X) = 32X 6 − 48X 4 + 18X 2 − 1
T3(X) = 4X 3 − 3X T7(X) = 64X 7 − 112X 5 + 56X 3 − 7X
T4(X ) = 8X 4 − 8X 2 + 1 Recurrence: TN (X ) = 2X · TN−1(X ) − TN−2(X)
Alternatively, we can write:

TN (X) = COS[N.ACS(X)] for |X| ⭐ 1 (6.17)


TN (X) = COSH[N.ACSH(X)] for |X| > 1 (6.18)
Figure 6.9 shows the Chebyshev polynomials up to N = 4.
From Figure 6.9 and the definition we can observe the following properties:

• TN (X) varies between −1 and +1 when the argument X is restricted to the range (−1, +1).
• TN (X = 0) = 0 if N is even.
• | TN (X = 0) | = 1 if N is odd.
• | TN (X) | grows toward infinity for | X | > 1, and the growing rate increases with N.

In the case of the Chebyshev filter response, the argument ω is positive and the equation
uses TN2 . Figure 6.10 shows this behaviour up to N = 5.
We can observe that for N odd, T N2 (X = 0) = 0 and we will have (N − 1)/2 additional passes
through zero in the pass band (0 < X < 1). Conversely, for N even, we have T N2 (X = 0) = 1
and N/2 passes through zero.
Finally, if we consider the total equation for the Chebyshev attenuation, Figure 6.11
shows this behaviour for N = 3 and N = 4.
388 Microwave Devices, Circuits and Subsystems for Communications Engineering

T N (X)

+1

N=1
3
X
–1 +1
4
2
–1

Figure 6.9 Chebyshev polynomials to N = 4

T N 2(X)

N=2 3 4 5

+1

X
0
+1

Figure 6.10 Chebyshev filter response up to N = 5

From Figure 6.11 and the definition, we can extract the following properties:

• at ω = 0 Atn = 1 for N odd and Atn = 1 + K 2 for N even


• at ω = 1 Atn = 1 + K 2 for N even and odd. Atn | dB = RdB = 10 · Log(1 + K 2)
• The pass band has a constant ripple from zero to RdB. The pass band up to ω = 1 is called
BWR as opposed to BW3 that is the 3dB point
• at ω Ⰷ 1 Atn grows toward infinity as (K2/4) · (2ω)2N. This means that the growing rate
is 6N dB/octave but at a given frequency the stop band attenuation is higher than the
Butterworth characteristic.
Filters 389

Atn Atn(dB)

N=3 N=3
odd odd

RdB = 10.Log(1+K2)

BWR BWR

1+K 2 RdB

ω ω
1 1
1 rad/s 1 rad/s

Atn Atn(dB)

N=4 N=4
even even

RdB = 10.Log(1+K2)

BWR BWR

1+K 2 RdB

ω ω
1 1
1 rad/s 1 rad/s

Figure 6.11 Chebyshev filter attenuation for N = 3 and 4

In order to normalise this filter, it is necessary to know the frequency for which the attenuation
is exactly 3dB (Atn = 2). This occurs at a frequency named ω3dB slightly higher than 1 rad/
sec and depends on the order N and the ripple factor K:

Atn = 2 = 1 + K 2 · T 2N (ω3dB) → TN (ω3dB) = 1/K

and using the property TN(X) = COSH[N.ACSH(X)], we can find:

ω3dB = COSH[1/N.ACSH(1/K)] (6.19)

It is now evident that an attenuation function of the form:

Atn = 1 + K 2 · TN2 (ω3dB · ω) Atn | dB = 10 · Log[1 + K 2 · T N2 (ω3dB · ω)] (6.20)

ensures that, for any value of N and K, the attenuation is 3 dB for ω = 1. So we have
compatibility with the Butterworth response. Figure 6.12 shows these two responses for a
N = 4 filter.
In the second case, the maximum frequency for RdB is exactly:

ω RdB = 1/ω 3dB (6.21)


390 Microwave Devices, Circuits and Subsystems for Communications Engineering

Atn Atn

2 2
Atn = 1 + K 2 .T N (ω ) Atn = 1 + K 2 .T N (ω 3dB. ω )
N=4 N=4
even even
BW3 BW3

2 2

BWR BWR

1+K 2 1+K 2

ω ω
1 1
1 1
ω3dB ωRdB

Figure 6.12 Chebyshev LPF responses for N = 4

6.3.3 The Bessel Response


The Butterworth characteristic has a good behaviour in amplitude selectivity along with
an acceptable phase response. This means that its transient characteristic is normally good
for many applications. Conversely, the Chebyshev family (depending on the ripple factor K)
offers a very important increase in stop band attenuation but a poor phase response.
The Bessel response has been optimised in order to have a maximally flat phase response
in the passband. This means that the amplitude selectivity in the stop band has been seriously
degraded, but the filter response under transient operation or complex modulations is optimum.
The generic low pass transfer function that has a constant delay can be written as a ratio
of hyperbolic trigonometric functions of the frequency:

1
T ( s) = s = jω (6.22)
sinh(s) + cosh(s)

The expansion of the above equation is rather tedious and, in an approximate way, we can
write the attenuation as:

Atn | dB ≅ 3 · ω 2 for any N and valid up to ω = 2 (6.23)

This filter is again an ‘all pole filter’ and Figure 6.13 shows its frequency response. It is clear
that for ω > 2, the selectivity in the stop band increases with N and the linearity of the phase
response extends towards higher frequencies. In fact, these kinds of filters are only used
when the transient properties are critical.
As the mathematical filter response is not easy to use, Figure 6.14 shows in a graphical
form the filter response in the pass band and stop band for N up to 7.

6.3.4 The Elliptic Response


All the characteristics studied above are ‘all pole’, that is, infinite attenuation (transmission
zero) is obtained for infinite ω. In these cases the best frequency selectivity in the stopband
is assured by the Chebyshev response.
Filters Atn(dB) 391

N
BW3

ω
0
1 rad/s 2 rad/s

Figure 6.13 Bessel LPF response

Atn(dB)
Pass band
3

2.5

2
N
1.5

0.5

0
0 0,2 0,4 0,6 0,8 1
ω (rad/s)
Atn(dB)
Stop band
80

70 7
60 6
50 5

40 4

30 3
20 N=2

10

0
1 2 3 4 5 6
ω (rad/s)

Figure 6.14 Bessel LPF response for N up to 7


392 Microwave Devices, Circuits and Subsystems for Communications Engineering

G T (dB)
Pass band Stop band
0 N = 3 filters

Butterworth

Chebyshev

Elliptic
ω

Figure 6.15 Elliptic LPF response compared with the Butterworth and Chebyshev

Atn(dB)

Amin

N=5

RdB

0 ω
1 ωS

Figure 6.16 Definitions for the Elliptic filter

The Elliptic response allows transmission zeros at controlled finite frequencies. This is
accomplished by having a Chebyshev-like ripple RdB in the pass band and an extra ripple
in the stop band. This last property means that the transition from the pass band to the stop
band will be more abrupt than the Chebyshev response as shown in Figure 6.15.
Figure 6.16 shows the primary definitions of an Elliptic filter where:

• RdB is the ripple in the pass band.


• Amin is the minimum attenuation (ripple like) in the stop band.
• ω S is the lowest frequency where Amin occurs.

As we can see, the stop band region has transmission zeros at finite frequencies, and the first
one occurs at a frequency slightly greater than ωS. The attenuation of these kinds of filters
can be written as:

Atn = 1 + K2 · ZN2 (ω) (6.24)


Filters 393

Atn(dB)

θ=50°

40°

30°
N=3
RdB

0 ω
1

Figure 6.17 Variation of filter response with θ

where K corresponds to the pass band ripple RdB, and ZN is an Elliptic function of order N:

W ⋅ ( A22 − ω 2 ) ⋅ ( A42 − ω 2 ) . . . ( Am2 − ω 2 )


ZN (ω ) = N odd: m = (N − 1)/2 (6.25)
(1 − A22ω 2 ) ⋅ (1 − A42ω 2 ) . . . (1 − Am2 ω 2 )

( A22 − ω 2 ) ⋅ ( A42 − ω 2 ) . . . ( Am2 − ω 2 )


ZN (ω ) = N even: m = N/2 (6.26)
(1 − A22ω 2 ) ⋅ (1 − A42ω 2 ) . . . (1 − Am2 ω 2 )

The transmission zeros are at A2, A4, . . . Am and the poles are at the symmetric points 1/A2,
1/A4, . . . , 1/Am. This symmetry results in a controlled ripple in the pass band as well as in
the stop band. Straightforward calculations give us the zero and pole position as a function
of the pass band ripple factor K and the order N.
The last parameter associated with this kind of filter is the modulation angle θ that is
defined as:

θ = ASIN(1/ω S) with 0 < θ < 90° (6.27)

and is the most popular parameter that defines the filter selectivity. As θ increases, the posi-
tion of ω S decreases towards ω = 1, and the transition from the pass band to the stop band
is more abrupt with a lower value of Amin. This fact is shown in Figure 6.17. On the other
hand, if we have fixed values for θ (or their equivalent ω S) and the order N, the stop band
attenuation Amin can be increased by allowing a higher value of RdB in the pass band.
It is clear that it is not easy to draw the attenuation characteristics of Elliptic filters
because we have three independent variables: K, N and Amin. So we will defer this problem
to the next sections.

6.4 Low Pass Prototype Filter Design


The actual specifications for a given filter with a particular characteristic are always trans-
lated, in a first step, to the concept of low pass prototype filter. The generic ladder network
394 Microwave Devices, Circuits and Subsystems for Communications Engineering

(a) R

L1 L3 LN-1
N even 1 ohm
C2 C4 CN
Γin, Z in

L1 L3 LN
N odd 1 ohm
C2 C4 CN-1
Γin, Z in

(b) 1/R

L2 L4 LN
N even 1 ohm
C1 C3 CN-1
Γin, Z in

1/R

L2 L4 LN-1
N odd 1 ohm
C1 C3 CN
Γin, Z in

Figure 6.18 Prototype low pass filter definitions

that conforms to this kind of prototype filter is shown in Figure 6.18(a), where the load
resistor is always unity while the source resistor is in general R.
The dual network of the above description is shown in Figure 6.18(b), where the element
values Lj and Cj for any value of j are the same and the source resistor is inverted. The
frequency response of both networks is exactly the same.
If we look for any of the networks shown in Figure 6.18(a), the reflection coefficient at
the input, measured with respect to the reference R, is:

Zin − R
Γin = ρ = |Γ| (6.28)
Zin + R
Filters 395

On the other hand, the attenuation of this lossless network can be written as:

1 1
Atn = Pin / PL = = =
1 − ρ 2 1 − | ( Zin − R)/( Zin + R) | 2
1
= = . . . after some manipulation =
| Zin − R | 2
1−
( Zin + R)( Z *in + R)

| Zin − R | 2
Atn = 1 + (6.29)
2 R ⋅ ( Zin + Z in*)
in an analogous form, the above expression can be written as a function of the input admitt-
ance of the filter:

| Yin − G | 2
Atn = 1 + G = 1/R (6.30)
2G ⋅ (Yin + Y in*)
The general idea is to fit these expressions to the mathematical filter characteristics such as
Butterworth, Chebyshev, etc.
For the above prototype network, the circuit is reduced to that shown in Figure 6.19 at
zero frequency (DC).
In this case, Zin = 1 and the attenuation is:

(1 − R)2
Atn = 1 + for ω = 0 (DC) (6.31)
4R

6.4.1 Calculations for Butterworth Prototype Elements


We remember that the maximally flat characteristic in amplitude is:

Atn = 1 + (ω)2N Atn | dB = 10 · Log[1 + (ω)2N ] N = order number (6.32)


and must be fitted to the characteristic of the prototype network:

| Zin − R | 2
Atn = 1 + (6.33)
2 R ⋅ ( Zin + Z in*)

Circuit for DC 1 ohm

Figure 6.19 PLPF at DC


396 Microwave Devices, Circuits and Subsystems for Communications Engineering

L
1 ohm
C

Γin, Z in N=2

Figure 6.20 PLPF for order 2

So, the problem is to find the reactive element values that make both expressions identical.
For clarification purposes, we can develop the calculations for N = 2. In this case, the
prototype network can be reduced, for example, to Figure 6.20.
The Butterworth attenuation for N = 2 is:

Atn = 1 + ω 4 (6.34)

and the input impedance of the prototype network is:

1
Zin = jLω + (6.35)
1 + jCω

After substitution of Zin into the general expression for the attenuation of the prototype
network, we have:

(1 − R)2 + ω 2 ⋅ ( L2 + C 2 R2 − 2 LC) + ω 4 ⋅ L2C 2


Atn = 1 + (6.36)
4R

and using this equation we can observe that at DC (ω = 0) we have:

Atn(ω = 0) = 1 + (1 − R)2/4R for R ≠ 1 (mismatch at DC)

Atn(ω = 0) = 1 for R = 1 (matching at DC)

So we can illustrate the attenuation for both cases in Figure 6.21. We observe a linear
translation on the attenuation axis.
In the general case, for any R, the attenuation at DC and cut-off are respectively:

Atn(ω = 0) = 1 + (1 − R)2/4R for any R (6.37)

Atn(ω = 1: cut-off) = 2 · [1 + (1 − R)2/4R] for any R (3 dB point) (6.38)

If we compare the Butterworth attenuation for N = 2 and the mathematical expression for the
prototype ladder network, we can deduce that the term in ω 2 must be zero:

L2 + C2R2 − 2LC = 0 (1° condition) (6.39)


Filters 397

Atn Atn Butterworth


Butterworth N=2
N=2 R≠1
R=1 BW3

BW3

2.[1 + (1-R)2/4R]

2 1 + (1-R)2 /4R
ω ω
1
1rad/s 1rad/s

Circuit for DC 1 ohm

Figure 6.21 Attenuation characteristics of Butterworth PLPF

Furthermore, the attenuation at ω = 1 (cut-off ) must be 2 · [1 + (1 − R)2/4R]. This means the


following identity:

(1 − R)2 L2C 2
2 · [1 + (1 − R)2/4R] = 1 + + → LC = 1 + R (2° condition) (6.40)
4R 4R

From the above two conditions we can deduce the values of L and C:

1+ R (1 + R) ⋅ R2
C2 = ⋅ [1 ± (1 − R2 )1/2 ] L2 = (6.41)
R2 [1 ± (1 − R2 )1/2 ]

As an example, we can deduce the element values for two cases: R = 1 and R = 0.5:

R=1 C = 1.414 F L = 1.414 H Atn(ω = 0) = 1.0 (0 dB)


Atn(ω = 1) = 2.0 (3 dB)

R = 0.5 C = 3.346 F L = 0.448 H Atn(ω = 0) = 1.125 (0.5 dB)


Atn(ω = 1) = 2.25 (3.5 dB)

and the two resulting networks are shown in Figure 6.22.


From the equations for L and C we can observe that there is no solution for R > 1. It is
possible, however, to design filters with generator impedances greater than load impedances.
If we look for the filter R = 0.5, we can remember that the dual network can be designed as
in Figure 6.23.
So all the problem consists of is in generating dual networks. We must remember at this
point that if R ≠ 1 the attenuation at DC is not zero but [1 + (1 − R)2/4R].
398 Microwave Devices, Circuits and Subsystems for Communications Engineering

R=1

1.414 H
1
1.414 F
N=2

R=0.5

0.448 H
1
3.346 F
N=2

Figure 6.22 Example Butterworth PLPFs

R=2

3.346 H
1
0.448 F
N=2

Figure 6.23 Dual network for R = 0.5

There are recursive equations for the prototype values for any value of N. In the particular
case of R = 1 (equality of generator and load impedances) the filter values are symmetrical.
Table 6.1 (see Appendix) shows the prototype element values in Henries and Farads up to
N = 7 (sufficient for most applications) and the table is organised as follows:

1. If we read the element values using the template shown in the top of the table, the
associated network is shown at the upper part of Figure 6.24.
2. If we read the element values using the template shown in the bottom of the table, the
associated network is shown in the bottom of Figure 6.24 (dual network).

Obviously, from Table 6.1 we can deduce the two above filters calculated for N = 2.
Figure 6.25(a) shows as an example an N = 4 Butterworth prototype filter for R = 1 (read
from the top template of Table 6.1), while Figure 6.25(b) shows its dual network.
In the same way, Figure 6.26(a) shows an N = 3 Butterworth prototype filter for R = 2.5
(read from the bottom template of Table 6.1), while Figure 6.26(b) shows its dual network.
The attenuation values (in the pass band (ω < 1) or in the stop band (ω > 1) can be easily
calculated, at any frequency and for any order N, by using the Butterworth expression for the
Filters 399

L2 LN
N even 1
CN
C1 C3 N odd

N R C1 L2 C3 LN or CN

N 1/R L1 C2 L3 CN or LN

1/R

L1 L3 LN
N odd 1
CN
C2 N even

Figure 6.24 Network format for Table 6.1

(a) 1 (b) 1

1.848 H 0.765 H 0.765 H 1.848 H


1 1
0.765 F 1.848 F 1.848 F 0.765 F

Figure 6.25 Prototype Butterworth LPF with N = 4

(a) 2.5 (b) 0.4

1.425 H 4.064 H 0.604 H


1 1
0.604 F 1.425 F 4.064 F

Figure 6.26 Prototype Butterworth LPF with N = 3 and R = 2.5


400 Microwave Devices, Circuits and Subsystems for Communications Engineering

Atn(dB) Atn(dB)
Butterworth
24 Butterworth
18.88
N=4
N=3
R=1
R= 0.4 or 2.5
BW3
BW3
3.88
3 ω 0.88 ω
0
1 rad/s 2 rad/s 1 rad/s 2 rad/s

Figure 6.27 Attenuations for Butterworth filters with N = 3 and 4

attenuation. The only restriction is that for R ≠ 1 the theoretical characteristic is shifted by
an amount given by the attenuation at ω = 0, that is 1 + (1 − R)2/4R. Figure 6.27 shows the
predicted attenuations for the N = 4 and N = 3 filters of the examples.

6.4.2 Calculations for Chebyshev Prototype Elements


We remember that the mathematical Chebyshev characteristic is:

Atn = 1 + K 2 · TN 2(ω) → for ω = 1 Atn = 1 + K 2 (ripple point)

or alternatively:

Atn = 1 + K 2 · TN 2(ω3dB · ω) → for ω = 1 Atn = 2 (3dB point)

We are going to perform the same extraction process as Butterworth for N = 2, and we can
start by using the first mathematical characteristic: attenuation at cut-off is 1 + K 2:

Atn = 1 + K 2 · T 22 (ω) = 1 + K 2[1 − 4ω 2 + 4ω 4 ] (6.42)

and the ladder network attenuation is:

(1 − R)2 + ω 2 ⋅ ( L2 + C 2 R2 − 2 LC) + ω 4 ⋅ L2C 2


Atn = 1 + (6.43)
4R

By fitting both equations at ω = 0 we have:

1 + K 2 = 1 + (1 − R)2/4R (6.44)

and this means that in this case R is a function of the ripple factor K:

R = 2K 2 + 1 ± [4K 2 · (1 + K 2)]1/2 (6.45)

if we want the Atn = 1 + K 2 at DC. In general, R cannot be unity (equal source and load
resistance) for N even. In the case of N odd, we can have equality.
Filters 401

On the other hand, by equalling the general attenuation equations, we have:

(1 − R)2 ω 2 ⋅ ( L2 + C 2 R2 − 2 LC) ω 4 ⋅ L2C 2


1 + K 2 [1 − 4ω 2 + 4ω 4 ] = 1 + + + (6.46)
4R 4R 4R

or, using the condition for R:

( L2 + C 2 R2 − 2 LC) L2C 2
− 4 K 2ω 2 + 4 K 2ω 4 = ω 2 ⋅ + ω4 ⋅ (6.47)
4R 4R

that is, we arrive at a system of equations:

L2 + C2R2 − 2LC = −16K 2R (1° condition) (6.48)

L2C2 = 16K 2R (2° condition) (6.49)

to calculate L and C.
As an example for this N = 2 filter, we can suppose that the allowed ripple in the passband
is RdB = 0.1 dB:

10 · Log(1 + K 2) = 0.1 → K = 0.15262

and, as we have N even, there are two solutions different from the unity for the source
resistor R:

Ra = 1.35536
with Ra = 1/Rb
Rb = 0.73781

If our choice is, for example, R = Rb = 0.73781, then the solutions for the reactive elements are:

R = 0.73781 C = 0.843 F L = 0.622 H Atn(W = 0) = 1 + K 2 = 0.02329 (0.1 dB)


Atn(W = 1) = 1 + K 2 = 0.02329 (0.1 dB)

Figure 6.28 shows this filter along with its theoretical response calculated from the Chebyshev
formula.

R=0.7378 Atn
N=2
0.622 H
1 (0.1dB)
0.843 F
N=2 1+K2 ω
1
1

Figure 6.28 Chebyshev PLPF with N = 2 and R = 0.73781


402 Microwave Devices, Circuits and Subsystems for Communications Engineering

R=1.3553 Atn
N=2
0.843 H
1 (0.1dB)
0.622 F
N=2 1+K2 ω
1
1

Figure 6.29 Chebyshev PLPF with N = 2 and R = 1.35536

Atn Atn(dB)
N N
even even
10.Log[1+(1-R)2/4R]
[1+(1-R)2/4R]/[1+K2]
Rdb
2
1+(1-R) /4R

ω ω
1 0
1 1

Atn N Atn(dB) N
odd odd

[1+(1-R)2/4R].[1+K2]
Rdb

1+(1-R)2/4R
ω ω
1 0
1 1
10.Log[1+(1-R)2/4R]

Figure 6.30 Chebyshev PLPF responses with N = 2

Conversely, if our choice is R = Ra = 1.35536, then the L and C values result in an


imaginary form. This means that this ladder network is only possible for R < 1. Again we
can use the concept of dual network to obtain a filter with these characteristics and with a
generator resistance Ra, as shown in Figure 6.29.
If we now try to design the same filter for a source resistor R different to Ra or Rb (the
optimum resistors in order to have the ripple value 1 + K 2 at cut-off), we can still use
the system of equations but in this case the Chebyshev characteristics are shifted in the
attenuation axis as shown in Figure 6.30 for N even and odd, that is, always the DC point
should have Atn = 1 + (1 − R)2/4R while maintaining the ripple.
As an example, suppose that the desired source resistor is R = 0.5. In this case the system
of equations gives the following results:

R = 0.5 L = 0.288 H C = 1.5715 F


Filters 403

R=0.5

0.288 H Atn(dB) N=2


1
9.29
1.5715 F
N=2
3.41
R=2 0.1 dB

0.511
1.5715 H 0.411
ω
1 0
1 1.94 3
0.288 F
N=2

Figure 6.31 Chebyshev PLPF with N = 2, R = 0.5 and 2.0

and the dual for R = 1/0.5 = 2 is:


R=2 L = 1.5715 H C = 0.288 F
in both cases, the DC attenuation is 10 · Log[1 + (1 − R)2/4R] = 0.511 dB.
Figure 6.31 shows both circuits along with the theoretical response. We can observe the
curve is shifted by an amount equal to 0.511 − 0.1 = 0.411 dB.
Using the Chebyshev formulas we can know that ω3dB = 1.9432 and the attenuation at
this point should be 3 + 0.411 = 3.411 dB, and the attenuation at say ω = 3 is 8.88 + 0.411
= 9.29 dB.
In conclusion, Chebyshev filters with N even cannot have the same source and load
resistor if we want to have the ripple RdB at cut-off and there is an optimum value of R for
this purpose. Conversely, for N odd, the source and load resistors should be the same (R = 1)
if we want RdB at cut-off. Finally, if we choose any other resistor R different from the above
explained, the response of attenuation in decibel will be shifted by an amount given by:
10 · Log[1 + (1 − R)2/4R] for N odd (6.50)
10 · Log[1 + (1 − R)2/4R] − RdB for N even (6.51)
Up to now we have designed Chebyshev filters where ω = 1 corresponds to the end of the
ripple. In many applications, however, the designer needs to speak in terms of the 3dB
points. So it should be interesting to modify the above designs in order to ensure that ω = 1
corresponds to the 3dB point. In this case the attenuation characteristic is given by:
Atn = 1 + K 2 · T 2N (ω 3dB · ω) (6.52)
where ω 3dB = COSH[1/N.ACSH(1/K)] (6.53)
One can repeat the calculation process by moving the variable ω towards ω · ω3dB. In fact,
it is easy to demonstrate that it is enough to multiply the reactive elements L, C calculated in
the sense ω = 1 → RdB, by the factor ω3dB to have ω = 1 → 3dB.
404 Microwave Devices, Circuits and Subsystems for Communications Engineering

R = 0.7378 Atn(dB)
N=2
0.622 H
3
1
0.843 F
N=2 0.1 ω
0
1 ω3dB
1.943

Figure 6.32 Chebyshev PLPF with N = 2

R = 0.7378 Atn(dB)
N=2
1.2087 H
3
1
1.638 F
N=2 0.1 ω
0
1/ω3dB 1

Figure 6.33 Chebyshev LPLF with N = 2

As an example, Figure 6.32 repeats the filter design shown in Figure 6.28 for RdB = 0.1,
N = 2, R = 0.73781 and ω3dB = 1.9432 rad/s.
Figure 6.33 shows the modification of the filter for ω = 1 → 3dB.
The same simple process is valid for any other Chebyshev filter.
Table 6.2 (see Appendix) shows the Chebyshev prototype element values in Henries and
Farads up to N = 7 and for ripples ranging from 0.01 up to 3dB. The elements are calculated
in order to have RdB at ω = 1, that is, the R values for N even are the optimum, and unity
for N odd. The conventions are the same as for Butterworth filters.
Tables 6.3 to Table 6.7 (see Appendix) show the prototype elements for a variety of
source resistors, including the optima, and ripples up to N = 7. In these tables the point ω =
1 corresponds to an Atn = 3dB. So it is very easy to translate the reactive elements in this
tables in order to have ω = 1 → RdB simply by division of the elements by ω3dB. We must
remember at this point that if R is not the optimum one, the theoretical curve is shifted in the
attenuation axis.

6.4.3 Calculations for Bessel Prototype Elements


The same approach can be taken for a given Bessel characteristic of order N. As we know,
the Bessel characteristic can be approximated by:

Atn | db ≅ 3 · ω 2 for any N and valid up to ω = 2

or the graphic format for ω > 2 and N up to 7.


Filters 405

As the theoretical calculations are rather tedious, Table 6.8 (see Appendix) shows the
PLPF elements for this response. We must remember that the cut-off ω = 1 corresponds to
the 3dB point, and the details are the same as for the Butterworth response.

6.4.4 Calculations for Elliptic Prototype Elements


As these kinds of filters are not ‘all pole’ filters and they exhibits transmission zeros at finite
frequencies, the conventional ladder network is not appropriate in this case. Figure 6.34
shows the conventional Elliptic networks for N up to 7 as well as the number of transmission
zeros NZ.
We must remember that these filters have a pass band ripple of RdB such that ω = 1 →
RdB, and follow the Chebyshev criteria for the source resistor R. Furthermore, as can be
expected, the parallel LC combinations as well as the series LC in the dual network are
resonant at the transmission zeros.
The Elliptic PLPF elements are tabulated in Tables 6.9 to 6.16 (see Appendix) as a
function of the modulation angle, and for N up to 7. The pass band ripple is 0.01 dB or
0.18 dB which correspond to a reflection coefficient of 0.05 and 0.2 respectively.
Figure 6.35 shows the conventions for reading these tables. The upper network is valid for
the upper template and conversely for the dual network shown at the bottom. The values A2,
A4, . . . are the position of the transmission zeros.
As an example, Figure 6.36 shows an N = 3, RdB = 0.18 Elliptic PLPF having ωs = 2,
Amin = 26.5 dB and the transmission zero at A2 = 2.27. The generator and load impedances
are unity. Furthermore, Figure 6.37 shows an N = 4, RdB = 0.18 Elliptic PLPF having
ωs = 3, Amin = 56.8 dB, A2 = 3.28 and R = 0.66 using the dual network.

6.5 Filter Impedance and Frequency Scaling


We must remember that when designing prototype low pass filters with specified filter
characteristics, the cut-off is at ω = 1 rad/s (0.159 Hz) for an attenuation in general of 3dB,
and for a load resistor unity. It is clear that it is not usual to use the above impedance
level nor the frequency values, so it is necessary to scale the filter in order to meet actual
specifications.

6.5.1 Impedance Scaling


If we know the design of a low pass prototype filter having RL = 1 and any RG, it should be
interesting to derive another filter such that the loading resistors have the actual values while
maintaining the same filter shape.
A typical prototype filter ladder network is shown in Figure 6.38, and we will develop the
process for a N = 4 filter. The extrapolation for any value of N will be clear.
The input impedance of this network can be written as:

1
Zin = jω ⋅ L1 + (6.54)
1
jω ⋅ C2 +
1
jω ⋅ L3 +
jω ⋅ C 4 + 1
406 Microwave Devices, Circuits and Subsystems for Communications Engineering

N Networks for Elliptic PLPF Dual Network NZ

3 1

4 1

5 2

N Networks for Elliptic PLPF Dual Network NZ

6 2

7 3

Figure 6.34 Elliptic networks for N = 3 to 7


Filters 407

LN
R N even
L2
C2
1
C1 C3 CN
N odd

θ ωs Amin A2 A4 ... C1 C2 L2 C3 ... LN or CN

Φ ωs Amin A2 A4 ... L1 L2 C2 L3 ... CN or LN

1/R L1 L3

LN
L2 N odd

1
CN
N even
C2

Figure 6.35 Convention for Elliptic filter tables

Atn(dB)

0.9612H
26.5
R=1

N=3
0.2019F 1
1.0512F 1.0512F

0.18

0
ω
1 2 2,27

Figure 6.36 Elliptic PLPF with N = 3


408 Microwave Devices, Circuits and Subsystems for Communications Engineering

R=0.66 1.198H 1.881H


Atn(dB)

0.07689H 56.8

1 N=4
0.8477F
1.205F
0.18

0 ω
1 3 3,28

Figure 6.37 Elliptic filter with N = 4

L1 L3
1
C2 C4
Z in

Figure 6.38 Prototype filter ladder network

RG

L’1 L’3
RL
C’2 C’4
Z’ in

Figure 6.39 Filter network with arbitrary load resistance

Now we have to analyse another network in Figure 6.39 with different reactances and a
general load resistor RL.
For this network the input impedance is:

1
Z in′ = jω ⋅ L′1 + (6.55)
1
jω ⋅ C′2 +
RL
jω ⋅ L′3 +
jω ⋅ RL ⋅ C′4 + 1

If we want both networks to have the same frequency response, the following condition
should be verified:

Zin = Z′in /RL (6.56)


Filters 409

RG K.R G
L,C L’ = K.L
C’ = C/K
RL K.R L
Linear Linear
Two-Port Two-Port

Figure 6.40 Impedance transformation

by applying this condition to both expressions for the input impedance we arrive at:

C′2 = C2/RL L′1 = L1 · RL (6.57)

C′4 = C4/RL L′3 = L3 · RL (6.58)

and this result can be expanded to any number of sections. In general, any linear network
maintains its frequency response if all the resistances and inductors are multiplied by a
constant and the capacitors are divided by the same constant. This fact is illustrated for
our case in Figure 6.40.
Figure 6.41 shows this transformation for a N = 3 Butterworth prototype filter when this
filter will be used in a 50 ohm system:

1H
1
2F 2F

50

50 H
50
40mF 40mF

Figure 6.41 Impedance transformation for Butterworth filter

Remember that in the above transformations, the frequency characteristics remain unchanged,
that is cut-off ω = 1 corresponds to the 3dB point or the RdB point depending on the filter
characteristic.
410 Microwave Devices, Circuits and Subsystems for Communications Engineering

6.5.2 Frequency Scaling


In practical low pass filters, cut-off frequencies are never FC = 0.159 Hz, and most systems
use high pass, band pass and band stop filters with specified cut-offs and bandwidths. Thus,
it is necessary to develop frequency translations in order to derive these kinds of practical
filters from the well-known prototype low pass filter.
In the following calculations we assume that the load resistor is unity RL = 1 (the imped-
ance scaling is a different problem from frequency scaling and is normally done at the end
of the design) and we denote with ‘prime’ values the prototype low pass filter elements and
frequency, while ‘unprimed’ values are for the scaled filter. The cut-off frequency for the
low pass prototype will be ω ′ = 1 rad/s.

6.5.3 Low Pass to Low Pass Expansion


This expansion is used to derive a new filter where the actual cut-off frequency is ωC while
maintaining the same filter shape:

Pass band Attenuation


Prototype filter: 0 → ω′ = 1 Atn′ = Atn′(ω ′)
Actual low pass filter: 0 → ωC Atn = Atn(ω)

Using the following frequency transformation:

ω′ = 0 → ω = 0
ω ′ = ω /ω C (6.59)
ω′ = 1 → ω = ωC

both filters will have the same frequency response if the reactances and susceptances in both
filters are the same in its own frequency axis:

jX′(ω ′) = jX(ω) jB′(ω ′) = jB(ω) (6.60)

This fact is depicted in Figure 6.42.


By applying the above statements we have:

X′ = L′ · ω ′ ≡ X = L · ω → L · ωC · ω′ = L · ω → L = L′/ω C (6.61)

B′ = C′ · ω ′ ≡ B = C · ω → C · ω C · ω ′ = C · ω → C = C′/ω C (6.62)

L’1 L’3 L1 L3
1 1
C’2 C’4 C2 C4
Low Pass Prototype Filter @ 1 Low Pass Filter @ω C
(ω’) (ω)

Figure 6.42 LPF frequency transformation


Filters 411

that is:

Prototype Low Pass


Inductor: L′ → Inductor: L with L = L′/ω C
Capacitor: C′ → Capacitor: C with C = C′/ω C

In conclusion, it is sufficient to divide all the prototype inductors and capacitors by ω C to


have the same low pass behaviour with a new cut-off frequency ω C.
Figure 6.43 shows the frequency translation for a N = 3 Chebyshev filter with ω ′ = 1 →
RdB. We observe the same shape with an expansion in the frequency axis.
As an example, suppose we have an N = 3 Chebyshev prototype filter with R = 1 and
RdB = 0.1, and we wish to have a low pass filter with a cut-off frequency of 200 MHz
such as FC → RdB. Using the tables, the prototype filter is shown in Figure 6.44(a). As
ω C = 2 · π · 200 · 106, all the inductors and capacitors should be divided by this value. The
resulting filter is shown in Figure 6.44(b).
If we want, we can do an impedance transformation at this stage to have a 200 MHz low
pass filter in, for example, a 50 ohm system as shown in Figure 6.45(a) and Figure 6.45(b).
Remember that 3dB point is at ω3dB = 1.388, that is, at 277.8 MHz.

Atn Atn

Chebyshev
Chebyshev
N=3
N=3

2 2

1+K 2 1+K 2

ω’ ω
–1 +1 –ω C +ω C
Low Pass Prototype Filter @ 1 Low Pass Filter @ ω C
(ω’) (ω)

Figure 6.43 Chebyshev filter frequency translation

Chebyshev N = 3 RdB = 0.1 R = 1 Chebyshev N = 3 RdB = 0.1 R = 1


R=1 1.1474 H R=1 0.9131nH

1 1
1.0315 F 1.0315 F 0.8208nF 0.8208nF
Low Pass Prototype Filter @ 1 Low Pass Filter @ Fc = 200MHz
(ω’) (ω)
(a) (b)

Figure 6.44 Chebyshev filter frequency translation example


412 Microwave Devices, Circuits and Subsystems for Communications Engineering

Atn(dB)

Chebyshev
Chebyshev N = 3 RdB = 0.1 R = 50
N=3
R = 50 45.65 nH
3

50
16.42 pF 16.42 pF
0.1
F(MHz)
Low Pass Filter @ FC = 200MHz 0
(ω) 200 277.8
Low Pass Filter @ FC = 200MHz
(a) (b)

Figure 6.45 LPF frequency and impedance transformation

The expected attenuation at, for example, 600 MHz can be easily calculated. At this
frequency we have:
ω ′ = ω /ω C = F/FC = 600 MHz / 200 MHz = 3
and using the PLPF Chebyshev characteristic for N = 3 and RdB = 0.1 (K 2 = 0.02329), we
can write:
Atn | dB = 10 · Log[1 + K 2 · T 23 (ω ′)] = 23.6 dB

6.5.4 Low Pass to High Pass Transformation


This transformation is used to derive a high pass filter where the actual cut-off frequency is
ωC while maintaining the same filter shape as the prototype low pass:

Pass band Attenuation


Prototype filter: 0 → ω′ = 1 Atn′ = Atn′(ω ′)
Actual high pass filter: ωC → ∞ Atn = Atn(ω)

Using the following frequency transformation:


ω′ = 0 → ω=∞
ω ′ = −ω C /ω ω ′ = +1 → ω = −ω C (6.63)
ω ′ = −1 → ω = +ω C
both filters will have the same frequency response if we use a high pass topology and if the
reactances and susceptances in both filters are the same in their own frequency axis. This
fact is depicted in Figure 6.46.
By applying the identities we have:
X′ = L′ · ω ′ = L′ · (−ω C /ω) ≡ X = −1/Cω → C = 1/(L′ · ω C) (6.64)
B′ = C′ · ω ′ = C′ · (−ω C /ω) ≡ B = −1/Lω → L = 1/(C′ · ω C) (6.65)
Filters 413

C1 C3

L’1 L’3
1 1
C’2 C’4 L2 L4
Low Pass Prototype Filter @ 1 High Pass Filter @ ω C
(ω’) (ω)

Figure 6.46 LP to HP transformation

Atn Atn

Chebyshev
Chebyshev
N=3
N=3

2 2

1+K2 1+K2

ω’ ω
–1 +1 –ω C +ω C
Low Pass Prototype Filter @ 1 High Pass Filter @ ω C
(ω’) (ω)

Figure 6.47 LP to HP Chebyshev filter transformation

that is, the inductors translate to capacitors and vice versa.

Prototype High Pass


Inductor: L′ → Capacitor: C with C = 1/(L′ · ω C)
Capacitor: C′ → Inductor: L with L = 1/(C′ · ω C)

Figure 6.47 shows this frequency transformation for an N = 3 Chebyshev filter with ω ′ = 1
→ RdB. We observe the same shape expanded in the frequency axis and symmetrical with
respect to ω = 0.
As an example, suppose we want to design an N = 3 Chebyshev high pass filter having
a ripple of RdB = 0.1 and a 3dB cut-off frequency of 400 MHz. The input and output
impedances should be 25 and 50 ohm respectively (R = 25/50 = 0.5). Figure 6.48(a) and
6.48(b) show the frequency translation while Figure 6.49(a) and 6.49(b) show the final filter
with impedance scaling and the filter response. We should remember that the mismatch due
to R = 0.5 is 0.511 dB.
The expected attenuation at, for example, 200 MHz can easily be calculated. At this
frequency, we have:

ω ′ = −ω C /ω = −FC /F = −400 MHz / 200 MHz = −2


ω3dB = 1.389.
414 Microwave Devices, Circuits and Subsystems for Communications Engineering

(a) Chebyshev N = 3 RdB = 0.1 R = 0.5 (b) Chebyshev N = 3 RdB = 0.1 R = 0.5
0.5 0.5

0.8383 H 474.6 pF
1 1
1.8530 F 3.1594 F 0.2147nH 0.1259nH
Low Pass Prototype Filter @ 1 High Pass Filter @ F C = 400MHz
(ω’) (ω)

Figure 6.48 LP to HP Chebyshev filter transformation example

(a) (b) Atn(dB)


Chebyshev N = 3 RdB = 0.1 R = 25 Chebyshev
N=3
25

3.51
9.493 pF
50 0.1 dB
10.74 nH 6.297 nH 0.61

High Pass Filter @ FC = 400MHz 0.51 F(MHz)


(ω) 400

High Pass Filter @ FC = 400 MHz

Figure 6.49 Chebyshev filter frequency translation example

Using the PLPF Chebyshev characteristic for N = 3 and RdB = 0.1 (K 2 = 0.02329) and the
ω3dB value, we can write:

Atn | dB = 10 · Log[1 + K 2 · T 23 (ω ′ · ω3dB)] = 21.48 dB + 0.511 = 21.99 dB

6.5.5 Low Pass to Band Pass Transformation


This transformation is used to derive a band pass filter where the actual cut-off frequencies
are ω 1 and ω 2 while maintaining the same filter shape as the prototype low pass:

Pass band Attenuation


Prototype filter: 0 → ω′ = 1 Atn′ = Atn′(ω ′)
Actual band pass filter: ω1 → ω2 Atn = Atn(ω)

Using the following frequency transformation:


ω0
ω′ = ⋅ [ω /ω 0 − ω 0 /ω ] where ω 0 = (ω 1 · ω 2)1/2
ω2 − ω1
or
ω 2 − ω1 1
ω = ω′ · ± [ω ′2(ω 2 − ω 1)2 + 4ω 1ω 2]1/2 (6.66)
2 2
Filters 415

L’1 L’3
1
C’2 C’4
Low Pass Prototype Filter @ 1
(ω’)

L1 L3
C1 C3

1
L2 C2 L4 C4
Band Pass Filter @ ω 1:ω 2
(ω)

Figure 6.50 LP to BP transformation

then we have:
ω ′ = 0 → ω = ±ω 0
ω ′ = +1 → ω = +ω 2 and ω = −ω 1
ω ′ = −1 → ω = −ω 2 and ω = +ω 1
both filters will have the same frequency response if we use a band pass topology and if the
reactances and susceptances in both filters are the same in their own frequency axis. This fact
is depicted in Figure 6.50 where each series inductor in the prototype translates to a series
LC resonant circuit, while each shunt capacitor moves to a parallel LC resonant network. All
the LC resonators (series or parallel) have their resonant frequency at ω 0 = (ω 1 · ω 2)1/2.
By applying the reactance identities we have:

ω0
Prototype series inductor: X ′ = L′ ⋅ ω ′ = L′ ⋅ ⋅ [ω /ω 0 − ω 0 /ω ]
ω 2 − ω1
Band pass LC series resonant: X = Lω − 1/Cω

that is,
X′ = X → L = L′/(ω 2 − ω 1) and C = 1/(L · ω 02)

in an analogous form we can derive:

ω0
Prototype parallel capacitor: B′ = C′ ⋅ ω ′ = C′ ⋅ ⋅ [ω /ω 0 − ω 0 /ω ]
ω 2 − ω1
Band pass LC parallel resonant: B = Cω − 1/Lω
416 Microwave Devices, Circuits and Subsystems for Communications Engineering

Atn Atn

Butterworth
N=3
Butterworth
N=3

2 2

ω’ ω
–1 +1 – ω2 –ω0 –ω 1 ω1 ω0 ω2
Low Pass Prototype Filter @ 1 Band Pass Filter @ ω 1: ω 2
(ω’) (ω)

Figure 6.51 Butterworth LP to BP transformation

that is,

B′ = B → C = C′/(ω 2 − ω 1) and L = 1/(C · ω 02 )

Prototype Band pass


Inductor: L′ → Series LC with L = L′/(ω 2 − ω 1) C = 1/(L · ω 02 )
Capacitor: C′ → Parallel LC with C = C′/(ω 2 − ω 1) L = 1/(C · ω 02 )
ω 02 = (ω 1 · ω 2)1/2 = LC

Figure 6.51 shows this frequency transformation for an N = 3 Butterworth filter with ω ′ = 1
→ 3dB. We observe the same shape expanded in the frequency axis as the low pass filter
and symmetrical with respect to ω = 0.
As an example we can design an N = 3 Butterworth band pass filter where the 3dB points
are at F1 = 820 MHz and F2 = 900 MHz and the generator and load impedances are respect-
ively RG = 100 ohm and RL = 50 ohm.
In this case F0 = 859.069 MHz and RG /RL = R = 2, and we know that this means that in
the prototype the attenuation at DC is Atn(ω ′ = 0) = 1 + (1 − R2)/4R = 1.125 (0.51 dB) and
the attenuation at cut-off is Atn(ω ′ = 1) = 2 *1.125 = 2.25 (3.51 dB). If we think about the
frequency translation, it is clear that in the actual band pass filter the attenuation at F0 will
be 0.51 dB and the attenuation at F1 and F2 will be 3.51 dB.
The prototype filter, from the tables, and the frequency transformation are shown in
Figure 6.52 for R = 2. Figure 6.53(a) and 6.53(b) show the final filter design, after impedance
scaling, and the predicted frequency response fom the Butterworth formula. In this case, we
have drawn the transfer gain GT in dB.
The expected attenuation at, for example, 700 MHz can easily be calculated. At this
frequency, we have:

ω0 F0
ω′ = ⋅ [ω /ω 0 − ω 0 /ω ] = ⋅ [ F / F0 − F0 / F] = − 4.4285
ω 2 − ω1 F0 − F1
Filters Butterworth N = 3 R = 2 417
2

1.181 H 3.261 H
1
0.779 F
Low Pass Prototype Filter @ 1
(ω’)

Butterworth N = 3 R = 2
2 2.349nH 14.608pF 6.487nH 5.2905pF

22.147pH

1.549nF
1

Band-Pass Filter @ 820:900 MHz


(ω)

Figure 6.52 Butterworth filter example of LP to BP frequency transformation

(a) Butterworth N = 3 R = 100

100 117.4nH 0.2921pF 324.3nH 0.1058pF


1.107nH

30.98pF

50

Band-Pass Filter @ 820:900 MHz


(ω)
(b) GT (dB)

0
–0.51 Butterworth
N=3
–3.51

F(MHz)
820 859 900
Band-Pass Filter @ 820:900 MHz

Figure 6.53 Butterworth filter example LP to BP impedance transformation


418 Microwave Devices, Circuits and Subsystems for Communications Engineering

and using the PLPF Butterworth characteristic for N = 3 we can write:


Atn | dB = 10 · Log[1 + (ω ′)2N ] = 38.77 dB + 0.511 = 39.28 dB
that is GT | dB = −39.28 dB.

6.5.6 Low Pass to Band Stop Transformation


This transformation is used to derive a band stop filter where the actual cut-off frequencies
are ω 1 and ω 2 while maintaining the same filter shape as the prototype low pass:
Pass band Attenuation
Prototype filter: 0 → ω′ = 1 Atn′ = Atn′(ω ′)
Actual band stop filter: 0 → ω 1 and ω 2 → inf Atn = Atn(ω)
Using the following frequency transformation:
ω2 − ω1 1
ω′ = ⋅ where ω 0 = (ω 1 · ω 2)1/2
ω0 [ω 0 /ω − ω /ω 0 ]
or
ω 2 − ω1 1
ω = (−1/ω′) · ± [(1/ω ′)2 · (ω 2 − ω 1)2 + 4ω 1ω 2 ]1/2 (6.67)
2 2
then we have:
ω ′ = 0 → ω = ± ω0
ω ′ = +1 → ω = +ω 1 and ω = −ω 2
ω ′ = −1 → ω = −ω 1 and ω = +ω 2
both filters will have the same frequency response if we use a band stop topology and if the
reactances and susceptances in both filters are the same in ther own frequency axis. This fact
is depicted in Figure 6.54 where each series inductor in the prototype translates to a parallel
LC resonant circuit, while each shunt capacitor moves to a series LC resonant network. All
the LC resonators (series or parallel) have their resonant frequency at ω 0 = (ω 1 · ω 2)1/2.
By applying the reactance identities, we have:
ω 2 − ω1 1
Prototype series inductor: X ′ = L′ ⋅ W ′ = L′ ⋅

ω0 [ω 0 /ω − ω /ω 0 ]
Band stop LC parallel resonant: X = −1/[Cω − 1/Lω]
that is,
X′ = X → L = L′ · (ω 2 − ω 1)/ω 20 and C = 1/(L · ω 02 )
in an analogous form we can derive:
ω 2 − ω1 1
Prototype series inductor: B′ = C′ ⋅ ω ′ = C′ ⋅ ⋅
ω0 [ω 0 /ω − ω /ω 0 ]
Band stop LC series resonant: B = −1/[Lω − 1/Cω]
Filters 419

L’1 L’3
1
C’2 C’4
Low Pass Prototype Filter @ 1
(ω’)

L1 L3
C1 C3
L2 L4

C2 C4

Band Stop Filter @ 0: ω 1 and ω 2 :∞


(ω)

Figure 6.54 Low pass to band stop tranformation

that is,

B′ = B → C = C′ · (ω 2 − ω 1)/ω 02 and L = 1/(C · ω 02 )

Prototype Band stop


Inductor: L′ → Parallel LC with L = L′ · (ω 2 − ω 1)/ω 02 C = 1/(L · ω 02)
Capacitor: C′ → Series LC with C = C′ · (ω 2 − ω 1)/ω 02 L = 1/(C · ω 02)
ω 0 = (ω 1 · ω 2) = LC
2 1/2

Figure 6.55 shows this frequency transformation for an N = 3 Butterworth filter with ω ′ = 1
→ 3dB. We observe the same shape expanded in the frequency axis as the high pass filter
and symmetrical with respect to ω = 0.
As an example, we design an N = 3 RdB = 0.1 band stop Chebyshev filter where the
Rdb points are at F1 = 5 GHz and F2 = 7 GHz and the generator and load impedances are
50 ohm.
In this case F0 = 5.916 GHz and RG /RL = R = 1. This means that in the prototype low pass
filter the attenuation at DC is 0 dB, the attenuation at cut-off is RdB = 0.1 and the filter is
symmetrical. Using the frequency translation the attenuation at F1 and F2 will be RdB, the
attenuation at F0 will be infinite and the ripple is maintained when going towards DC and
high frequency.
420 Microwave Devices, Circuits and Subsystems for Communications Engineering

Atn Atn
Butterworth
Butterworth N=3
N=3

2 2

ω’ ω
–ω 2 –ω 0 –ω 1 ω1 ω0 ω2
–1 +1

Figure 6.55 Butterworth LP to BS filter example, frequency transformation

Chebyshev N = 3 RdB = 0.1 R = 1


1

1.1474 H
1
1.0315 F 1.0315 F

Low Pass Prototype Filter @ 1

Chebyshev N = 3 RdB = 0.1 R = 1

1
10.435 pH

69.35 pF
77.14 pH 77.14 pH

9.381 pF 9.381 pF

Band Stop Filter @ 0 : 5 GHz and 7 : ∞ GHz

Figure 6.56 Chebyshev LP to BS filter example, frequency translation

The prototype filter, from the tables, and the frequency transformation are shown in
Figure 6.56 for R = 1. Figures 6.57(a) and 6.57(b) show the final filter design, after impedance
scaling, and the predicted frequency response fom the Chebyshev formula.
The expected attenuation at, for example, 6.5 GHz can easily be calculated. At this
frequency, we have:
Filters 421

(a) Chebyshev N = 3 RdB = 0.1 R = 50

50
0.5217 nH

1.387 pF
3.857 nH 3.857 nH

50

0.1876 pF 0.1876 pF

Band Stop Filter @ 0 : 5 GHz and 7 : ∞ GHz

(b) Atn(dB)

Chebyshev
N=3

0.1
F(GHz)
0
5.0 5.91 7.0
Band Stop Filter @ 0 : 5 GHz and 7 : ∞ GHz

Figure 6.57 Chebyshev LP to BS filter example, impedance transformation

F2 − F1 1
ω′ = ⋅ = −1.7931
F0 [ F0 / F − F / F0 ]

and using the PLPF Chebyshev characteristic for N = 3 and RdB = 0.1 (K 2 = 0.02329) we
can write:

Atn | dB = 10 · Log[1 + K 2 · T 23 (ω ′)] = 9.18 dB

6.5.7 Resonant Network Transformations


As we will see in the next section, band pass and band stop translations for Elliptic filters
result in networks as shown in Figure 6.58(a), that is, a series resonant circuit at ω 0 in
parallel with a parallel LC resonant circuit at the same frequency ω 0.
422 Microwave Devices, Circuits and Subsystems for Communications Engineering

(a) ω0 (b)
ωa ωb
L1 1/(ω 0 2L1)
La Lb

1/(ω 0 2C1)
Ca Cb
C1
ω0

Figure 6.58 Band pass to band stop filter transformation

It should be very interesting to find an equivalent, shown in Figure 6.58(b), where we


have a cascade combination of two parallel LC networks. It can easily be shown that both
networks are equivalent if:

1
La = Lb = H · La
C1 ⋅ ω 20 ⋅ ( H + 1)
1 1
Ca = Cb =
H ⋅ ω 20 ⋅ La ω 20 ⋅ La

1
where H =1+ ⋅ {1 + [1 + 4 ⋅ ω 20 ⋅ L1 ⋅ C1]1/2} (6.68)
2 ⋅ L1 ⋅ C1 ⋅ ω 20

and the resonant frequencies of the two parallel circuits are respectively:

ωa = H 1/2 · ω 0 ω b = H −1/2 · ω 0 (6.69)

From the above equations it is easy to see that La,Cb and Lb,Ca resonate at the primitive
frequency ω 0.
The same approach can be applied to the circuits shown in Figure 6.59(a) and 6.59(b).
In this case, the equations are:

1 1
Ca = Cb =
L1 ⋅ ω 20 ⋅ ( H + 1) ω 20 ⋅ La
H +1 1
La = ⋅ L1 Lb =
H ω 20 ⋅ Ca

1
where H =1+ ⋅ {1 + [1 + 4 ⋅ ω 20 ⋅ L1 ⋅ C1]1/2} (6.70)
2 ⋅ L1 ⋅ C1 ⋅ ω 20

and the resonant frequencies of the two series circuits are respectively:

ωa = H 1/2 · ω 0 ω b = H −1/2 · ω 0 (6.71)


Filters 423

(a) (b) ωa
ω0
La Ca
L1 1/(ω 0 2L1)

1/(ω 0 2C1)

C1
ω0 Lb
Cb
ωb

Figure 6.59 Band stop to band pass filter transformation

6.6 Elliptic Filter Transformation


It is clear that all the developed formulas for impedance scaling and frequency transformations
are valid here, but there are some details in respect to the position of the transmission zeros
that are interesting to note. In order to clarify the situation we will develop the successive
transformations starting from an example:

N = 5 Elliptic PLPF with R = 1, RdB = 0.01, θ = 50°

This PLPF has ωs = 1.3054, Amin = 24.94 dB and two transmission zeros at A2 = 1.9480 and
A4 = 1.3481. The filter structure, as well as the dual, are shown in Figure 6.60 and the filter
response is shown in Figure 6.61. We can observe that the first parallel network resonates at
A2 while the second parallel network resonates at A4.

6.6.1 Low Pass Elliptic Translation

In this case, every inductor and capacitor is divided by the new cut-off frequency ωc. These
calculations, along with the filter response, are shown in Figure 6.62 for an LPF having a
Fc = 600 MHz. Every parallel LC network resonates at ωc times the zero position:

Elliptic N = 5 Rdb = 0.01 R = 1 θ = 50 Dual Network

A2 A4 0.5894H 1.2145H 0.2392H


0.2498H

0.9446H

1.0545H 0.5824H 1
1
0.2498F 0.9446F 1
1.0545F

0.5824F

1
0.5894F 1.2145F 0.2392F
Low Pass Prototype Filter @ 1 Low Pass Prototype Filter @ 1

Figure 6.60 Elliptic filter transformation example


424 Microwave Devices, Circuits and Subsystems for Communications Engineering

A2
A4
Atn(dB)

24.94

N=5

1.3481
0.01 1.948

0 ω’
1 1.3054

Figure 6.61 Response of filter shown in Figure 6.60

Elliptic N = 5 RdB = 0.01 R = 1 θ = 50


A2. ω c A4. ω c

0.2797nH 0.1545nH
1
66.26pF 250.56pF

1
156.34pF 322.16pF 66.45pF
Low Pass Filter @ 600 MHz

A2
A4
Atn(dB)

24.94

N=5

808.8 MHz
0.01 1168.8 MHz

0
F(MHz)
600 783.2

Figure 6.62 Elliptic LP filter transformation


Filters 425

ω ′ = ω s = 1.3054 → ω = 4921.24 106 rad/s → F = 783.24 MHz


ω ′ = A2 = 1.9480 → ω = 7343.78 106 rad/s → F = 1168.8 MHz
ω ′ = A4 = 1.3481 → ω = 5082.21 10 rad/s → F = 808.8 MHz
6

Impedance scaling should be performed at this point.

6.6.2 High Pass Elliptic Translation


Every capacitor translates to an inductor and vice versa. Figure 6.63 shows this transforma-
tion for a cut-off frequency of Fc = 600 MHz. As before, the position of the transmission
zeros follows the rules for the frequency translation:

ω ′ = ω s = 1.3054 → F = 459.63 MHz


ω ′ = A2 = 1.9480 → F = 308.00 MHz
ω ′ = A4 = 1.3481 → F = 445.07 MHz

Elliptic N = 5 Rdb = 0.01 R = 1 θ = 50


ωc /A2 ω c /A4

1
251.55pF 455.45pF

1.0618nH 0.2808nH 1
0.45nH 0.2184nH 1.1089nH
High Pass Filter @ 600 MHz

Atn(dB)
A2 A4

24.94

N=5

308 MHz
445.07 MHz
0.01

0
F(MHz)
459.6 600

Figure 6.63 Elliptic LP to HP filter transformation


426 Microwave Devices, Circuits and Subsystems for Communications Engineering

Elliptic N = 5 RdB = 0.01 R = 1 θ = 50


ω0 ω0

ω0 0.8391nH 86.245pF ω0 0.4364nH 156.15pF ω0


1
0.364nH 0.0963nH
0.1543nH

0.0748nH

0.3802nH
966.47pF

190.35pF
469.03pF

198.78pF 751.68pF 1

A2 A4
Band Pass Filter @ 500 MHz : 700 MHz

Figure 6.64 Elliptic LP to BP filter transformation

6.6.3 Band Pass Elliptic Translation


Here each parallel capacitor in the PLPF translates to a parallel LC network that resonates
at the centre frequency ω 0, and each series inductor translates to a series LC resonant circuit
at the same frequency ω 0. If we suppose that F1 = 500 MHz and F2 = 700 MHz (Fc =
591.6 MHz), the translated circuit is shown in Figure 6.64, where all the series or parallel
resonant circuits have their resonance at ω 0. The band pass translation formulas give:

ω ′ = ωs = 1.3054 → F = 736.37 MHz


→ F = 475.3 MHz
ω ′ = A2 = 1.9480 → F = 817.65 MHz (F2a)
→ F = 428.05 MHz (F2b)
ω ′ = A4 = 1.3481 → F = 741.58 MHz (F4a)
→ F = 471.96 MHz (F4b)

With this circuit, it is not easy, for example, to adjust the transmission zeros, that is, this
network does not meet the requirements of their positions. Using the transformations for the
resonant networks we obtain the circuit shown in Figure 6.65 where the parallel resonant
circuits in the series branches have their resonant frequencies at the transmission zero.
Furthermore, Figure 6.66 shows the frequency response of this BPF.

6.6.4 Band Stop Elliptic Translation


Here each inductance in the PLPF translates to a parallel LC network that resonates at
the centre frequency ω 0, and each capacitor translates to a series LC resonant circuit at the
same frequency ω 0. If we suppose as above that F1 = 500 MHz and F2 = 700 MHz (F0 =
591.6 MHz), the translated circuit is shown in Figure 6.67, where all the series or parallel
resonant circuits have their resonance at ω 0. The band stop translation formulas give:

ω ′ = ω s = 1.3054 → F = 519.89 MHz


→ F = 673.15 MHz
Filters 427

Elliptic N = 5 RdB = 0.01 R = 1 θ = 50


ω 2a ω 2b ω 4a ω 4b
ω0
ω0 ω0
0.125nH 0.239nH 0.0374nH 0.0588nH
1
302.8pF 578.5pF 1230pF 1932.8pF

96 6.47pF
0.0748nH
0.1543nH

0.3802nH

190.35pF
469.03pF

A2 A4
Band Pass Filter @ 500 MHz : 700 MHz

Figure 6.65 Elliptic LP to BP filter transformation including transmission zeros

Atn(dB)
A2 A2
A4 A4

24.94

N=5

471.96 MHz 741.58 MHz


428.05 MHz
0.01 817.65 MHz

0
F(MHz)
475.3 500 591.6 700 736.7

Figure 6.66 Frequency response of Elliptic BP filter

Elliptic N = 5 RdB = 0.01 R = 1 θ = 50


ω0 ω0

ω0 0.0959nH ω0 0.0529nH ω0
1
754.64pF 1366.3pF
0.655nH

3.327nH
1.350nH

3.185nH 22.72pF 0.8424nH 85.907pF


1
110.45pF

21.754pF
53.603pF

A2 A4
Band Stop Filter @ 500 MHz : 700 MHz

Figure 6.67 Elliptic LP to BS transformation


428 Microwave Devices, Circuits and Subsystems for Communications Engineering

ω ′ = A2 = 1.9480 → F = 645.16 MHz (F2a)


→ F = 542.49 MHz (F2b)
ω ′ = A4 = 1.3481 → F = 670.41 MHz (F4a)
→ F = 522.06 MHz (F4b)

Again, with this circuit, it is not easy to adjust the transmission zeros. Using the transformations
for the resonant networks we obtain the circuit shown in Figure 6.68 where the parallel resonant
circuits in the series branches have their resonant frequencies at the transmission zeros.
Furthermore, Figure 6.69 shows the frequency response of this BSF.

Elliptic N = 5 RdB = 0.01 R = 1 θ = 50


ω 2a ω 2b ω 4a ω 4b
ω0
1
0.0438nH 0.0521nH 0.0232nH 0.0298nH

1389.2pF 1652.08pF 2430.26pF 3120.84pF


1.350nH

0.655nH

3.327nH
ω0 1
A2 A4
53.603pF

110.45pF

21.754pF
ω0
Band Stop Filter @ 500 MHz : 700 MHz

Figure 6.68 Elliptic BS filter including transmission zeros

Atn(dB)
522.06 MHz

645.16 MHz
670.41 MHz
542.49 MHz

591.6 MHz

24.94

N=5

0.01

0
F(MHz)
500 519.8 673.1 700

Figure 6.69 Frequency response of Elliptic BS filter


Filters 429

6.7 Filter Normalisation


When we are involved in a filter design process we not only have specifications in the
pass band but most of the time also in the stop band. As we know, these stop band require-
ments will impose the number of sections as well as the filter response. It seems clear now
that is imperative to accurately translate the actual filter specifications to PLPF constraints.
For this purpose it is convenient to write here the frequency translation equations:

Prototype Low Pass to Low Pass:

ω ′ = ω /ω C (6.72)

Prototype Low Pass to High Pass:

ω ′ = −ω C /ω (6.73)

Prototype Low Pass to Band Pass:

ω0
ω′ = ⋅ [ω /ω 0 − ω 0 /ω ] where ω0 = (ω 1 · ω 2)1/2
ω 2 − ω1
or

ω 2 − ω1 1 2
ω = ω′ ⋅ ± [ω ′ (ω 2 − ω 1 )2 + 4ω 1ω 2 ]1/2 (6.74)
2 2

Prototype Low Pass to Band Stop:

ω 2 − ω1 1
ω′ = ⋅ where ω 0 = (ω 1 · ω 2)1/2
ω0 [ω 0 /ω − ω /ω 0 ]

or

ω 2 − ω1 1
ω = (−1/ω ′) ⋅ ± [(1/ω ′)2 ⋅ (ω 2 − ω 1 )2 + 4ω 1ω 2 ]1/ 2 (6.75)
2 2

where ω C is the cut-off frequency in low pass and high pass filters, while ω 1 and ω 2 are the
cut-off frequencies in band pass and band stop filters.

6.7.1 Low Pass Normalisation


In this kind of filter it is common to specify the cut-off frequency FC (usually the 3dB point)
and a minimum of attenuation XdB at a given frequency Fa in the stop band, as shown in
Figure 6.70. Taking into account the frequency translation of a low pass filter, it is clear that
the frequency Fa has a PLPF response given by:

ω ′ = ω a /ω C = Fa /FC (6.76)

and with this response the PLPF should have the same attenuation as the original LPF, as
shown in Figure 6.70.
430 Microwave Devices, Circuits and Subsystems for Communications Engineering

GT(dB) GT (dB)
Pass band Stop band Pass band Stop band
0 0
-3 -3

LPF PLPF

-X -X

F ω’
Fc Fa 1 Fa /Fc

Figure 6.70 LPF normalisation

GT(dB) GT (dB)
Pass band Stop band Pass band Stop band
0 0
-3 -3

LPF PLPF

-40 -40

F(MHz) ω’
200 800 1 4

Figure 6.71 Example of LPF normalisation

As an example, suppose we have the following low pass specification:

• cut-off 3dB point at FC = 200 MHz


• 40 dB minimum attenuation at Fa = 800 MHz.

The value of ω ′ is 4 and the PLPF specifications should be:

• cut-off 3dB point at ω ′ = 1 rad/s


• 40 dB minimum attenuation at ω ′ = 4 rad/s.

This normalisation is depicted in Figure 6.71. We are now able to work with the PLPF
network design, performing the frequency and impedance scaling.

6.7.2 High Pass Normalisation


In this case the specifications are the same as low pass filters, that is, a cut-off frequency FC
and a desired attenuation of XdB at Fa. After the frequency translation equation for HPF,
the PLPF response is given by:

ω ′ = −ω C /ω a = −FC /Fa (6.77)


Filters 431

GT(dB) GT(dB)
Stop band Pass band Pass band Stop band
0 0
-3 -3

HPF PLPF

-X
-X
F ω’
Fa Fc 1 Fc /Fa

Figure 6.72 HPF transform and normalisation

GT(dB) GT(dB)
Stop band Pass band Pass band Stop band
0 0
-3 -3

HPF PLPF

-50
-50
F(MHz) ω’
100 200 1 2

Figure 6.73 Example of HPF normalisation

The PLPF shown in Figure 6.72 should have XdB of attenuation. But this frequency
translation is symmetrical with respect to ω ′ = 0, so it is more convenient to look for a PLPF
frequency given by the negative of the above:
ω ′ = +ω C /ω a = +FC /Fa (6.78)
As an example, suppose we have the following high pass specification:

• cut-off 3dB point at FC = 200 MHz


• 50 dB minimum attenuation at Fa = 100 MHz.

ω ′ = 2 and consequently the PLPF specification, shown in Figure 6.73, should be:

• cut-off 3dB point at ω ′ = 1 rad/s


• 50 dB minimum attenuation at ω ′ = 2 rad/s.

6.7.3 Band Pass Normalisation


In the case of BPF, Figure 6.74, we can distinguish between broadband and narrowband
BPF depending on the fractional bandwidth FBW given by:
FBW = (F2 − F1)/F0 = BW/F0 (6.79)
432 Microwave Devices, Circuits and Subsystems for Communications Engineering

GT(dB) BPF
Stop band Pass band Stop band
0
-3
BW

-X

F
Fa F1 F0 F2 Fb

Figure 6.74 BPF response

Filters having a FBW ≥ 0.707 (70.7%) are considered broadband filters. This limit is given
by the fact that in this case F2 = 2F1, that is, we have an octave of bandwidth.
In the general case, this BPF has specifications at the cut-off frequencies F1, F2 and at one
or two stopband frequencies Fa, Fb.

6.7.3.1 Broadband band pass normalisation


When we deal with a BPF having broadband characteristics, the filter implementation is
done by using a cascade connection of an LPF and an HPF as shown in Figure 6.75.
The LPF and HPF sections are normalised independently of the PLPF as shown in
Figure 6.76.
As an example suppose we have the following broadband BPF specifications:

• cut-off RdB = 0.1 points at F1 = 4.5 GHz and F2 = 10 GHz


• 40 dB minimum attenuation at Fa = 2 GHz and Fb = 20 GHz.

LPF HPF

GT(dB) GT(dB)
Stop band Pass band Stop band
0 0
-3 -3
LPF
HPF
BPF BPF

-X -X

F F
Fa F1 F2 Fb Fa F1 F2 Fb

Figure 6.75 BPF generated from LPF and HPF


Filters 433

GT(dB) GT(dB)

0 0
-3 -3

LPF PLPF

-X -X

F ω’
F2 Fb 1 Fb /F2

GT(dB) GT(dB)

0 0
-3 -3

HPF PLPF

-X -X

F ω’
Fa F1 1 F1/Fa

Figure 6.76 LPF and HPF normalisations

Then the PLPF specifications for the LPF section are:

• cut-off RdB = 0.1 at ω ′ = 1 rad/s


• 40 dB minimum attenuation at ω ′ = 2 rad/s.

And for the HPF section are:

• cut-off RdB = 0.1 at ω ′ = 1 rad/s


• 40 dB minimum attenuation at ω ′ = 2.25 rad/s.

The two normalisations are shown in Figure 6.77. The next step is to design both PLPF and
then to perform the two frequency translations.

6.7.3.2 Narrowband band pass normalisation


These BPF have a bandwidth less than one octave and in this case we can directly use the
frequency translation properties. It is convenient to take into account that such a frequency
translation, Figure 6.78, is symmetrical in a geometric form with respect to F0 = (F1 · F2)1/2,
that is, a given attenuation of XdB at a given frequency Fx is also encountered at its sym-
metrical Fxs = F 02 /Fx.
434 Microwave Devices, Circuits and Subsystems for Communications Engineering

GT(dB) GT(dB)

0 0
-0.1 -0.1

LPF PLPF

-40 -40

F(GHz) ω’
10 20 1 2

GT(dB) GT(dB)

0 0
-0.1 -0.1

HPF PLPF

-40 -40

F(GHz) ω’
2 4.5 1 2.25

Figure 6.77 Example LPF and HPF normalisations

GT(dB)
BPF
0
-3

-X

F
Fx F1 F0 F2 Fxs

Figure 6.78 Narrowband BPF response

In effect, the PLPF frequency for Fx is:

F0
ω ′x = ⋅ [ Fx / F0 − F0 / Fx ] (6.80)
F2 − F1

while the corresponding value for Fxs is:


F0 F0
ω ′xs = ⋅ [ Fxs / F0 − F0 / Fxs ] = ⋅ [ F0 / Fx − Fx / F0 ] = −ω s′ (6.81)
F2 − F1 F2 − F1
Filters 435

GT(dB) GT(dB)
BPF BPF
0 Specifications: 0 Specifications:
-3 Xdb at Fa and Fb -3 XdB at Fa and Fb

Fas < Fb Fas > Fb

-X -X

Retain Retain
F F
Fbs Fa F0 Fas Fb Fa Fbs F0 Fb Fas

Figure 6.79 Non-symmetric BPF responses

and we know that the attenuation characteristics of the different PLPF are the same if the
frequency changes sign.
Most of the time the actual stop band requirements XdB at the two stop band frequencies
Fa and Fb are such that Fa and Fb are not symmetrical with respect to F0. So we have to
convert these specifications into a symmetrical form. This is accomplished by obtaining the
symmetry Fas = F 20 /Fa. If Fas < Fb this means that at Fb we will have more than XdB and we
will retain the stop band symmetrical bandwidth (Fa : Fas). Conversely if Fas > Fb, then the
attenuation at Fb will be less than XdB and we will retain the stop band symmetrical
bandwidth (Fbs : Fb) because it is clear that in this case Fa < Fbs. Both cases are shown in
Figure 6.79.
As an example we have the following narrowband BPF specifications:

• 3dB bandwidth BW = 300 MHz centered around 1 GHz


• 64 dB minimum attenuation at ± 300 MHz.

In this case F1 = 850 MHz, F2 = 1150 MHz and F0 = 988.68 MHz with a FBW = 0.30 (30%).
Furthermore, the stop band frequencies are Fa = 700 MHz and Fb = 1300 MHz.
The value of Fas is 1396.4 MHz, that is, greater than Fb, so the retained stop band require-
ments are Fbs = 751.9 MHz and Fb = 1300 MHz. Now we have to translate Fbs or Fb into the
PLPF ω ′ plane using the equation for the BPF frequency translation. Both results should
be the same except for the sign. The resulting value is ω ′ = 1.827. As a consequence, the
PLPF specifications are:

• 3dB cut-off at ω ′ = 1 rad/s


• 64 dB minimum attenuation at ω ′ = 1.827 rad/s.

Figure 6.80 shows this normalization.

6.7.4 Band Stop Normalisation


As in the case of BPF, in the BSF – Figure 6.81 – we can distinguish between broadband
and narrowband BSF depending on the FBW.
436 Microwave Devices, Circuits and Subsystems for Communications Engineering

GT(dB)
BPF GT(dB)
0
-3
300 MHz 0
-3

-64 PLPF

-64
600 MHz
F(MHz) ω’
700 751.9 988.7 1300 1396.4 1 1.827
Fa Fbs F0 Fb Fas

Figure 6.80 Example BPF normalisation

GT(dB)
BW

Pass band Stop band Pass band


0
-3

BSF

-X

F
F1 Fa Fb F2

Figure 6.81 BSF definitions

6.7.4.1 Broadband band stop normalisation


In dealing with BPF having broadband characteristics, the filter implementation is done
by using a parallel connection of a LPF and a HPF with the output combined as shown in
Figure 6.82.
The LPF and HPF sections are normalised independently of the PLPF as shown in
Figure 6.83.
As an example, suppose we have the following broadband BSF specifications:

• cut-off 3dB points at F1 = 2 GHz and F2 = 8 GHz


• 40 dB minimum attenuation at Fa = 3 GHz and Fb = 5 GHz.

Then the PLPF specifications for the LPF section are:

• cut-off 3dB point at ω ′ = 1 rad/s


• 40 dB minimum attenuation at ω ′ = 1.5 rad/s.
Filters 437

LPF

HPF Σ

GT(dB) GT(dB)
BSF BSF
Pass band Stop band Pass band
0 0
-3 -3
LPF

-X -X
HPF
F F
F1 Fa Fb F2 F1 F a Fb F2

Figure 6.82 BSF realised by combination of LPF and HPF

GT(dB) GT(dB)

0 0
-3 -3

LPF PLPF

-X -X

F ω’
F1 Fa 1 Fa /F1

GT(dB) GT(dB)

0 0
-3 -3

HPF PLPF

-X -X

F ω’
Fb F2 1 F2 /Fb

Figure 6.83 LPF and HPF normalisations


438 Microwave Devices, Circuits and Subsystems for Communications Engineering

GT(dB) GT(dB)

0 0
-3 -3

LPF PLPF

-40 -40

F(GHz) ω’
2 3 1 1.5

GT(dB) GT(dB)

0 0
-3 -3

HPF PLPF

-40 -40

F(GHz) ω’
5 8 1 1.6

Figure 6.84 Example LPF and HPF normalisations

and for the HPF section are:

• cut-off 3dB point at ω ′ = 1 rad/s


• 40 dB minimum attenuation at ω ′ = 1.6 rad/s.

The two normalisations are shown in Figure 6.84. Remember that the two frequency transla-
tions are made independently.

6.7.4.2 Narrowband band stop normalisation


All the considerations for BPF are applicable here. So it is easiest to explain with an
example. Suppose we have the following BSF specifications:

• 3dB cut-off bandwidth BW = 400 MHz around 1 GHz


• 50 dB minimum attenuation at ± 100 MHz.

Here we have F1 = 800 MHz, F2 = 1200 MHz, F0 = 979.8 MHz with a FBW = 0.4 (40%).
The stop band frequencies are Fa = 900 MHz and Fb = 1100 MHz.
The value of Fas is 1066.6 MHz, that is less than Fb, and this means that the presumed
attenuation at Fb should be less than 50 dB. So the retained values are Fbs = 872.7 MHz and
Fb = 1100 MHz. Now we translate these values into the ω ′ plane by using the BSF frequency
translation. The result is ω ′ = 1.76 and the PLPF specifications should be:
Filters 439

GT(dB) GT(dB)
BSF

0 400 MHz 0
-3 -3
200 MHz
PLPF
-50
-50

F(GHz) ω’
.8 .87 .9 1.1 1.2 1 1.76
F1 Fbs Fa Fb F2

Figure 6.85 Narrowband BSF normalisation

• 3dB cut-off at ω ′ = 1 rad/s


• 50 dB minimum attenuation at ω ′ = 1.76 rad/s.

Figure 6.85 shows this normalisation.

Self-assessment Problems
6.1 Design a 3-stage low pass Butterworth filter with a cut-off frequency of 5 GHz
for use in a circuit with a normalised load of R = 1.

6.2 Design a 4-stage high pass Chebyshev filter with a cut-off frequency of 6 GHz
for use in a circuit with a normalised load of R = 2, RdB = 0.5.

6.3 Design a 2-stage band pass Bessel filter with cut-off frequencies of 1GHz and
3 GHz for use in a circuit with a normalised load of R = 1.

6.4 Design a 3-stage band stop Elliptical filter with cut-off frequencies of 2 GHz and
4 GHz for use in a circuit with a normalised load of R = 1, RdB = 0.01.
Appendix Tables for Filter Design 440

Table 6.1 Low pass prototype filter: Butterworth

N R C1 L2 C3 L4 C5 L6 C7 N R C1 L2 C3 L4 C5 L6 C7

2 1.000 1.414 1.414 5 1.000 0.618 1.618 2.000 1.618 0.618


1.111 1.035 1.835 0.900 0.442 1.027 1.910 1.756 1.389
1.250 0.849 2.121 0.800 0.470 0.866 2.061 1.544 1.738
1.429 0.697 2.439 0.700 0.517 0.731 2.285 1.333 2.108
1.667 0.566 2.828 0.600 0.586 0.609 2.600 1.126 2.552
2.000 0.448 3.346 0.500 0.686 0.496 3.051 0.924 3.133
2.500 0.342 4.095 0.400 0.838 0.388 3.736 0.727 3.965
3.333 0.245 5.313 0.300 1.094 0.285 4.884 0.537 5.307
5.000 0.156 0.707 0.200 1.608 0.186 7.185 0.352 7.935
10.00 0.074 14.814 0.100 3.512 0.091 14.095 0.173 15.710
Inf 1.414 0.707 Inf 1.545 1.694 1.382 0.894 0.309
3 1.000 1.000 2.000 1.000 6 1.000 0.518 1.414 1.932 1.932 1.414 0.518
0.900 0.808 1.633 1.599 1.111 0.289 1.040 1.322 2.054 1.744 1.335
0.800 0.844 1.384 1.926 1.250 0.245 1.116 1.126 2.239 1.550 1.688
0.700 0.815 1.165 2.277 1.429 0.207 1.236 0.957 2.499 1.346 2.062
0.600 1.023 0.965 2.702 1.667 0.173 1.407 0.801 2.858 1.143 2.509
0.500 1.181 0.779 3.261 2.000 0.141 1.653 0.654 3.369 0.942 3.094
0.400 1.425 0.604 4.064 2.500 0.111 2.028 0.514 4.141 0.745 3.931
0.300 1.838 0.440 5.363 3.333 0.082 2.656 0.379 5.433 0.552 5.280
0.200 2.669 0.284 7.910 5.000 0.054 3.917 0.248 8.020 0.363 7.922
0.100 5.167 0.138 15.455 10.00 0.026 7.705 0.122 15.786 0.179 15.738
Inf 1.500 1.333 0.500 Inf 1.553 1.759 1.553 1.202 0.758 0.259
4 1.000 0.765 1.848 1.848 0.765 7 1.000 0.445 1.247 1.802 2.000 1.802 1.247 0.445
1.111 0.466 1.592 1.744 1.469 0.900 0.299 0.711 1.404 1.489 2.125 1.727 1.296
1.250 0.388 1.695 1.511 1.811 0.800 0.322 0.606 1.517 1.278 2.334 1.546 1.652
1.429 0.325 1.862 1.291 2.175 0.700 0.357 0.515 1.688 1.091 2.618 1.350 2.028
1.667 0.269 2.103 1.082 2.613 0.600 0.408 0.432 1.928 0.917 3.005 1.150 2.477
2.000 0.218 2.452 0.883 3.187 0.500 0.480 0.354 2.273 0.751 3.553 0.951 3.064
2.500 0.169 2.986 0.691 4.009 0.400 0.590 0.278 2.795 0.592 4.380 0.754 3.904
3.333 0.124 3.883 0.507 5.338 0.300 0.775 0.206 3.671 0.437 5.761 0.560 5.258
5.000 0.080 5.684 0.331 7.940 0.200 1.145 0.135 5.427 0.287 8.526 0.369 7.908
10.00 0.039 11.094 0.162 15.642 0.100 2.257 0.067 10.700 0.142 16.822 0.182 15.748
Inf 1.531 1.577 1.082 0.383 Inf 1.558 1.799 1.659 1.397 1.055 0.656 0.223
Microwave Devices, Circuits and Subsystems for Communications Engineering

N 1/R L1 C2 L3 C4 L5 C6 L7 N 1/R L1 C2 L3 C4 L5 C6 L7

Note: (ω = 1 rad/s for Atn = 3dB)


Table 6.2 Low pass prototype filter: Chebyshev
Filters

N RdB R C1 L2 C3 L4 C5 N RdB R C1 L2 C3 L4 C5 L6 C7 C8 L9

2 0.01 1.1007 0.4077 0.4488 6 0.01 1.1007 0.7098 1.4970 1.5350 1.6896 1.3600 0.7813
0.1 1.3554 0.6220 0.8430 0.1 1.3554 0.8618 1.9029 1.5170 2.0562 1.4039 1.1681
0.2 1.5386 0.6745 1.0378 0.2 1.5386 0.8838 2.0974 1.4555 2.2394 1.3632 1.3598
0.5 1.9841 0.7071 1.4029 0.5 1.9841 0.8696 2.4758 1.3137 2.6064 1.2479 1.7254
1.0 2.6599 0.6850 1.8219 1.0 2.6599 0.8101 2.9367 1.1518 3.0634 1.1041 2.1546
2.0 4.0957 0.6075 2.4881 2.0 4.0957 0.6964 3.7151 0.9393 3.8467 0.9071 2.8521
3.0 5.8095 0.5339 3.1013 3.0 5.8095 0.6033 4.4641 0.7929 4.6061 0.7685 3.5045
3 0.01 1.0000 0.6291 0.9702 0.6291 7 0.01 1.0000 0.7969 1.3924 1.7481 1.6331 1.7481 1.3924 0.7969
0.1 1.0000 1.0315 1.1474 1.0315 0.1 1.0000 1.1811 1.4228 2.0966 1.5733 2.0966 1.4228 1.1811
0.2 1.0000 1.2275 1.1525 1.2275 0.2 1.0000 1.3722 1.3781 2.2756 1.5001 2.2756 1.3781 1.3722
0.5 1.0000 1.5963 1.0967 1.5963 0.5 1.0000 1.7372 1.2583 2.6381 1.3444 2.6381 1.2583 1.7372
1.0 1.0000 2.0236 0.9941 2.0236 1.0 1.0000 2.1664 1.1116 3.0934 1.1736 3.0934 1.1116 2.1664
2.0 1.0000 2.7107 0.8327 2.7107 2.0 1.0000 2.8655 0.9119 3.8780 0.9535 3.8780 0.9119 2.8655
3.0 1.0000 3.3487 0.7117 3.3487 3.0 1.0000 3.5182 0.7723 4.6386 0.8039 4.6386 0.7723 3.5182
4 0.01 1.1007 0.6476 1.3212 1.2003 0.7128 8 0.01 1.1007 0.7333 1.5554 1.6193 1.8529 1.6833 1.7824 1.4130 0.8072
0.1 1.3554 0.8180 1.7703 1.3061 1.1088 0.1 1.3554 0.8778 1.9444 1.5640 2.1699 1.6010 2.1199 1.4346 1.1897
0.2 1.5386 0.8468 1.9761 1.2844 1.3028 0.2 1.5386 0.8972 2.1349 1.4925 2.3413 1.5217 2.2963 1.3875 1.3804
0.5 1.9841 0.8419 2.3661 1.1926 1.6703 0.5 1.9841 0.8796 2.5093 1.3389 2.6964 1.3590 2.6564 1.2647 1.7451
1.0 2.6599 0.7892 2.8311 1.0644 2.0991 1.0 2.6599 0.8175 2.9685 1.1696 3.1488 1.1839 3.1107 1.1161 2.1744
2.0 4.0957 0.6819 3.6063 0.8806 2.7925 2.0 4.0957 0.7016 3.7477 0.9510 3.9335 0.9605 3.8948 0.9151 2.8733
3.0 5.8095 0.5920 4.3471 0.7483 3.4389 3.0 5.8095 0.6073 4.4990 0.8018 4.6990 0.8089 4.6575 0.7745 3.5277
5 0.01 1.0000 0.7563 1.3049 1.5773 1.3049 0.7563 9 0.01 1.0000 0.8144 1.4270 1.8043 1.7125 1.9057 1.7125 1.8043 1.4270 0.8144
0.1 1.0000 1.1468 1.3712 1.9750 1.3712 1.1468 0.1 1.0000 1.1950 1.4425 2.1345 1.6167 2.2053 1.6167 2.1345 1.4425 1.1956
0.2 1.0000 1.3394 1.3370 2.1660 1.3370 1.3394 0.2 1.0000 1.3860 1.3938 2.3093 1.5340 2.3728 1.5340 2.3093 1.3938 1.3860
0.5 1.0000 1.7058 1.2296 2.5408 1.2296 1.7058 0.5 1.0000 1.7504 1.2690 2.6678 1.3673 2.7239 1.3673 2.6678 1.2690 1.7504
1.0 1.0000 2.1349 1.0911 3.0009 1.0911 2.1349 1.0 1.0000 2.1797 1.1192 3.1215 1.1897 3.1747 1.1897 3.1215 1.1192 2.1797
2.0 1.0000 2.8310 0.8985 3.7827 0.8985 2.8310 2.0 1.0000 2.8790 0.9171 3.9056 0.9643 3.9598 0.9643 3.9056 0.9171 2.8790
3.0 1.0000 3.4817 0.7618 4.5381 0.7618 3.4817 3.0 1.0000 3.5340 0.7760 4.6692 0.8118 4.7272 0.8118 4.6692 0.7760 3.5340

N RdB 1/R L1 C2 L3 C4 L5 N RdB 1/R L1 C2 L3 C4 L5 C6 L7 C8 L9


441

Note: (ω = 1 rad/s for Atn = RdB)


Table 6.3 Low pass prototype filter: Chebyshev (RdB = 0.01)
442

N R C1 L2 C3 L4 C5 L6 C7 N R C1 L2 C3 L4 C5 L6 C7

2 1.1007 1.3472 1.4829 5 1.0000 0.9766 1.6849 2.0366 1.6849 0.9766


1.1111 1.2472 1.5947 0.9000 0.8798 1.4558 2.1738 1.6412 1.2739
1.2500 0.9434 1.9974 0.8000 0.8769 1.2350 2.3785 1.4991 1.6066
1.4286 0.7591 2.3442 0.7000 0.9263 1.0398 2.6582 1.3228 1.9772
1.6667 0.6091 2.7496 0.6000 1.0191 0.8626 3.0408 1.1345 2.4244
2.0000 0.4791 3.2772 0.5000 1.1658 0.6985 3.5835 0.9421 3.0092
2.5000 0.3634 4.0328 0.4000 1.3983 0.5442 4.4027 0.7491 3.8453
3.3333 0.2590 5.2546 0.3000 1.7966 0.3982 5.7721 0.5573 5.1925
5.0000 0.1642 7.6498 0.2000 2.6039 0.2592 8.5140 0.3679 7.8257
10.000 0.0781 14.749 0.1000 5.0406 0.1266 16.741 0.1819 15.613
Inf 1.4118 0.7415 Inf 1.5466 1.7950 1.6449 1.2365 0.4883
3 1.0000 1.1811 1.8214 1.1811 6 1.1007 0.8514 1.7956 1.8411 2.0266 1.6312 0.9372
0.9000 1.0917 1.6597 1.4802 1.1111 0.7597 1.7817 1.7752 2.0941 1.6380 1.0533
0.8000 1.0969 1.4431 1.8057 1.2500 0.5445 1.8637 1.4886 2.4025 1.5067 1.5041
0.7000 1.1600 1.2283 2.1653 1.4286 0.4355 2.0383 1.2655 2.7346 1.3318 1.8987
0.6000 1.2737 1.0236 2.5984 1.6667 0.3509 2.2978 1.0607 3.1671 1.1451 2.3568
0.5000 1.4521 0.8294 3.1644 2.0000 0.2786 2.6781 0.8671 3.7683 0.9536 2.9483
0.4000 1.7340 0.6452 3.9742 2.5000 0.2139 3.2614 0.6816 4.6673 0.7606 3.7899
0.3000 2.2164 0.4704 5.2800 3.3333 0.1547 4.2448 0.5028 6.1631 0.5676 5.1430
0.2000 3.1934 0.3047 7.8338 5.0000 0.0997 6.2227 0.3299 9.1507 0.3760 7.7852
0.1000 6.1411 0.1479 15.390 10.000 0.0483 12.171 0.1623 18.105 0.1865 15.595
Inf 1.5012 1.4330 0.5905 Inf 1.5510 1.8471 1.7897 1.5976 1.1904 0.4686
4 1.1000 0.9500 1.9382 1.7608 1.0457 7 1.0000 0.9127 1.5947 2.0021 1.8704 2.0021 1.5947 0.9127
1.1111 0.8539 1.9460 1.7439 1.1647 0.9000 0.8157 1.3619 2.0886 1.7217 2.2017 1.5805 1.2060
1.2500 0.6182 2.0749 1.5417 1.6170 0.8000 0.8111 1.1504 2.2618 1.5252 2.4647 1.4644 1.5380
1.4286 0.4948 2.2787 1.3336 2.0083 0.7000 0.8567 0.9673 2.5158 1.3234 2.8018 1.3066 1.9096
1.6667 0.3983 2.5709 1.1277 2.4611 0.6000 0.9430 0.8025 2.8720 1.1237 3.2496 1.1310 2.3592
2.0000 0.3156 2.9943 0.9260 3.0448 0.5000 1.0799 0.6502 3.3822 0.9276 3.8750 0.9468 2.9478
2.5000 0.2418 3.6406 0.7293 3.8746 0.4000 1.2971 0.5072 4.1563 0.7350 4.8115 0.7584 3.7900
3.3333 0.1744 4.7274 0.5379 5.2085 0.3000 1.6692 0.3716 5.4540 0.5459 6.3703 0.5682 5.1476
5.0000 0.1121 6.9102 0.3523 7.8126 0.2000 2.4235 0.2423 8.0565 0.3604 9.4844 0.3776 7.8019
10.000 0.0541 13.469 0.1729 15.510 0.1000 4.7006 0.1186 15.872 0.1784 18.818 0.1879 15.652
Inf 1.5287 1.6939 1.3122 0.5229 Inf 1.5593 1.8671 1.8657 1.7651 1.5633 1.1610 0.4564

N 1/R L1 C2 L3 C4 L5 C6 L7 N 1/R L1 C2 L3 C4 L5 C6 L7
Microwave Devices, Circuits and Subsystems for Communications Engineering

Note: (ω = 1 rad/s for Atn = 3dB)


Table 6.4 Low pass prototype filter: Chebyshev (RdB = 0.1)

N R C1 L2 C3 L4 C5 L6 C7 N R C1 L2 C3 L4 C5 L6 C7
Filters

2 1.3554 1.2087 1.6382 5 1.0000 1.3013 1.5559 2.2411 1.5559 1.3013


1.4286 0.9771 1.9824 0.9000 1.2845 1.4329 2.3794 1.4878 1.4883
1.6667 0.7326 2.4885 0.8000 1.2998 1.2824 2.5819 1.3815 1.7384
2.0000 0.5597 3.0538 0.7000 1.3580 1.1170 2.8679 1.2437 2.0621
2.5000 0.4169 3.8275 0.6000 1.4694 0.9469 3.2688 1.0846 2.4835
3.3333 0.2933 5.0502 0.5000 1.6535 0.7777 3.8446 0.9126 3.0548
5.0000 0.1841 7.4257 0.4000 1.9538 0.6119 4.7193 0.7333 3.8861
10.000 0.0868 14.433 0.3000 2.4765 0.4509 6.1861 0.5503 5.2373
Inf 1.3911 0.8191 0.2000 3.5457 0.2950 9.1272 0.3659 7.8890
0.1000 6.7870 0.1447 17.957 0.1820 15.745
Inf 1.5613 1.8069 1.7659 1.4173 0.6507
3 1.0000 1.4328 1.5937 1.4328 6 1.3554 0.9419 2.0797 1.6581 2.2473 1.5344 1.2767
0.9000 1.4258 1.4935 1.6219 1.4286 0.7347 2.2492 1.4537 2.5437 1.4051 1.6293
0.8000 1.4511 1.3557 1.8711 1.6667 0.5422 2.6003 1.1830 3.0641 1.1850 2.1739
0.7000 1.5210 1.1927 2.1901 2.0000 0.4137 3.0679 0.9575 3.7119 0.9794 2.7936
0.6000 1.6475 1.0174 2.6026 2.5000 0.3095 3.7652 0.7492 4.6512 0.7781 3.6453
0.5000 1.8530 0.8383 3.1594 3.3333 0.2195 4.9266 0.5514 6.1947 0.5795 4.9962
0.4000 2.1857 0.6603 3.9675 5.0000 0.1393 7.2500 0.3613 9.2605 0.3835 7.6184
0.3000 2.7630 0.4860 5.2788 10.000 0.0666 14.220 0.1777 18.427 0.1901 15.350
0.2000 3.9418 0.3172 7.8503 Inf 1.5339 1.8838 1.8306 1.7485 1.3937 0.6383
0.1000 7.5121 0.1549 15.466
Inf 1.5133 1.5090 0.7164
4 1.3554 0.9924 2.1476 1.5845 1.3451 7 1.0000 1.2615 1.5196 2.2392 1.6804 2.2392 1.5196 1.2615
1.4286 0.7789 2.3480 1.4292 1.7001 0.9000 1.2422 1.3946 2.3613 1.5784 2.3966 1.4593 1.4472
1.6667 0.5764 2.7304 1.1851 2.2425 0.8000 1.2550 1.2449 2.5481 1.4430 2.6242 1.3619 1.6967
2.0000 0.4398 3.2269 0.9672 2.8563 0.7000 1.3100 1.0826 2.8192 1.2833 2.9422 1.2326 2.0207
2.5000 0.3288 3.9605 0.7599 3.6976 0.6000 1.4170 0.9169 3.2052 1.1092 3.3841 1.0807 2.4437
3.3333 0.2329 5.1777 0.5602 5.0301 0.5000 1.5948 0.7529 3.7642 0.9276 4.0150 0.9142 3.0182
5.0000 0.1475 7.6072 0.3670 7.6143 0.4000 1.8853 0.5926 4.6179 0.7423 4.9702 0.7384 3.8552
10.000 0.0704 14.887 0.1802 15.229 0.3000 2.3917 0.4369 6.0535 0.5557 6.5685 0.5569 5.2167
Inf 1.5107 1.7682 1.4550 0.6725 0.2000 3.4278 0.2862 8.9371 0.3692 9.7697 0.3723 7.8901
0.1000 6.5695 0.1405 17.603 0.1838 19.376 0.1862 15.813
Inf 1.5748 1.8577 1.9210 1.8270 1.7340 1.3786 0.6307

N 1/R L1 C2 L3 C4 L5 C6 L7 N 1/R L1 C2 L3 C4 L5 C6 L7
443

Note: (ω = 1 rad/s for Atn = 3dB)


444

Table 6.5 Low pass prototype filter: Chebyshev (RdB = 0.25)

N R C1 L2 C3 L4 C5 L6 C7 N R C1 L2 C3 L4 C5 L6 C7

2 2 0.6552 2.7632 5 1 1.5046 1.4436 2.4050 1.4436 1.5046


3 0.3740 4.3118 0.5 3.0103 0.7218 3.6080 1.4436 1.5046
4 0.2637 5.7389 0.333 4.5149 0.4812 4.8100 1.4436 1.5046
8 0.1215 11.259 0.25 6.0196 0.3615 6.0130 1.4436 1.5046
Inf 1.3584 0.8902 0.125 12.040 0.1807 10.823 1.4436 1.5046
Inf 1.5765 1.7822 1.8225 1.4741 0.7523
3 1 1.6325 1.4360 1.6325 6 2 0.6867 3.2074 0.9308 3.8102 1.2163 1.7088
0.5 3.2663 1.0775 1.6325 3 0.4330 5.0976 0.5392 6.0963 1.0804 1.8393
0.333 4.8988 0.9572 1.6325 4 0.3173 6.9486 0.3821 8.2530 1.0221 1.8987
0.25 6.5326 0.8971 1.6325 8 0.1539 14.310 0.1762 16.719 0.9393 1.9868
0.125 13.064 0.8081 1.6325 Inf 1.5060 1.9221 1.8191 1.8329 1.4721 0.7610
Inf 1.5348 1.5285 0.8169
4 2 0.6747 3.6860 1.0247 1.8806 7 1 1.5120 1.4169 2.4535 1.5350 2.4535 1.4169 1.5120
3 0.4149 6.2744 0.7682 2.1302 0.5 3.024 0.7085 4.9069 1.1515 2.4535 1.4169 1.5120
4 0.3020 8.8161 0.6667 2.2533 0.333 4.5361 0.4723 7.3596 1.0230 2.4535 1.4169 1.5120
8 0.1448 19.020 0.5334 2.4516 0.25 6.0471 0.3542 9.8120 0.9593 2.4535 1.4169 1.5120
Inf 1.4817 1.8213 1.5068 0.7853 0.125 12.095 0.1776 19.625 0.8631 2.4535 1.4169 1.5120
Inf 1.6009 1.8287 1.9666 1.8234 1.8266 1.4629 0.7555

N 1/R L1 C2 L3 C4 L5 C6 L7 N 1/R L1 C2 L3 C4 L5 C6 L7

Note: (ω = 1 rad/s for Atn = 3dB)


Microwave Devices, Circuits and Subsystems for Communications Engineering
Table 6.6 Low pass prototype filter: Chebyshev (RdB = 0.5)

N R C1 L2 C3 L4 C5 L6 C7 N R C1 L2 C3 L4 C5 L6 C7
Filters

2 1.9841 0.9827 1.9497 5 1.0000 1.8068 1.3025 2.6914 1.3025 1.8068


2.0000 0.9086 2.1030 0.9000 1.8540 1.2220 2.8478 1.2379 1.9701
2.5000 0.5635 3.1647 0.8000 1.9257 1.1261 3.0599 1.1569 2.1845
3.3333 0.3754 4.4111 0.7000 2.0347 1.0150 3.3525 1.0582 2.4704
5.0000 0.2282 6.6995 0.6000 2.2006 0.8901 3.7651 0.9420 2.8609
10.000 0.1052 13.322 0.5000 2.4571 0.7537 4.3672 0.8098 3.4137
Inf 1.3067 0.9748 0.4000 2.8692 0.6091 5.2960 0.6640 4.2447
0.3000 3.5877 0.4590 6.8714 0.5075 5.6245
0.2000 5.0639 0.3060 10.054 0.3430 8.3674
0.1000 9.5560 0.1525 19.646 0.1731 16.547
Inf 1.6299 1.7400 1.9217 1.5138 0.9034
3 1.0000 1.8636 1.2804 1.8636 6 1.9841 0.9053 2.5774 1.3675 2.7133 1.2991 1.7961
0.9000 1.9175 1.2086 2.0255 2.0000 0.8303 2.7042 1.2912 2.8721 1.2372 1.9557
0.8000 1.9965 1.1203 2.2368 2.5000 0.5056 3.7219 0.8900 4.1092 0.8808 3.1025
0.7000 2.1135 1.0149 2.5172 3.3333 0.3370 5.0554 0.6323 5.6994 0.6348 4.4810
0.6000 2.2889 0.8937 2.8984 5.0000 0.2059 7.6145 0.4063 8.7319 0.4121 7.0310
0.5000 2.5571 0.7592 3.4360 10.000 0.0958 15.186 0.1974 17.681 0.2017 14.432
0.4000 2.9854 0.6146 4.2416 Inf 1.4618 1.9799 1.7803 1.9253 1.5077 0.8981
0.3000 3.7292 0.4633 5.5762
0.2000 5.2543 0.3087 8.2251
0.1000 9.8899 0.1534 16.118
Inf 1.5720 1.5179 0.9318
4 1.9841 0.9202 2.5864 1.3036 1.8258 7 1.0000 1.7896 1.2961 2.7177 1.3848 2.7177 1.2961 1.7896
2.0000 0.8452 2.7198 1.2383 1.9849 0.9000 1.8348 1.2146 2.8691 1.3080 2.8829 1.2335 1.9531
2.5000 0.5162 3.7659 0.8693 3.1205 0.8000 1.9045 1.1182 3.0761 1.2149 3.1071 1.1546 2.1681
3.3333 0.3440 5.1196 0.6208 4.4790 0.7000 2.0112 1.0070 3.3638 1.1050 3.4163 1.0582 2.4554
5.0000 0.2100 7.7076 0.3996 6.9874 0.6000 2.1744 0.8824 3.7717 0.9786 3.8524 0.9441 2.8481
10.000 0.0975 15.352 0.1940 14.262 0.5000 2.4275 0.7470 4.3695 0.8377 4.4886 0.8137 3.4050
Inf 1.4361 1.8888 1.5211 0.9129 0.4000 2.8348 0.6035 5.2947 0.6846 5.4698 0.6690 4.2428
0.3000 3.5456 0.4548 6.8674 0.5221 7.1341 0.5129 5.6350
0.2000 5.0070 0.3034 10.049 0.3524 10.496 0.3478 8.4041
0.1000 9.4555 0.1513 19.649 0.1778 20.631 0.1761 16.665
Inf 1.6464 1.7772 2.0306 1.7892 1.9239 1.5034 0.8948

N 1/R L1 C2 L3 C4 L5 C6 L7 N 1/R L1 C2 L3 C4 L5 C6 L7
445

Note: (ω = 1 rad/s for Atn = 3dB)


446

Table 6.7 Low pass prototype filter: Chebyshev (RdB = 1)

N R C1 L2 C3 L4 C5 L6 C7 N R C1 L2 C3 L4 C5 L6 C7

2 3 0.5723 3.1317 5 1 2.2072 1.1279 3.1025 1.1279 2.2072


4 0.3653 4.6002 0.5 4.4144 0.5645 4.6532 1.1279 2.2072
8 0.1571 9.6582 0.333 6.6216 0.3763 6.2050 1.1279 2.2072
Inf 1.2128 1.1093 0.25 8.8288 0.2822 7.7557 1.1279 2.2072
0.125 17.656 0.1406 13.961 1.1279 2.2072
Inf 1.7213 1.6448 2.0614 1.4928 1.1031
3 1 2.2160 1.0883 2.2160 6 3 0.6785 3.8725 0.7706 4.7107 0.9692 2.4060
0.5 4.4309 0.8168 2.2160 4 0.4810 5.6441 0.4759 7.3511 0.8494 2.5820
0.333 6.6469 0.7259 2.2160 8 0.2272 12.310 0.1975 16.740 0.7256 2.7990
0.25 8.8619 0.6799 2.2160 Inf 1.3775 2.0969 1.6896 2.0744 1.4942 1.1022
0.125 17.725 0.6120 2.2160
Inf 1.6522 1.4595 1.1080
4 3 0.6529 4.4110 0.8140 2.5346 7 1 2.2043 1.1311 3.1472 1.1942 3.1472 1.1311 2.2043
4 0.4517 7.0825 0.6118 2.8484 0.5 4.4075 0.5656 6.2934 0.8951 3.1472 1.1311 2.2043
8 0.2085 17.164 0.4275 3.2811 0.333 6.6118 0.3774 9.4406 0.7955 3.1472 1.1311 2.2043
Inf 1.3499 2.0102 1.4879 1.1057 0.25 8.8151 0.2828 12.588 0.7466 3.1472 1.1311 2.2043
0.125 17.631 0.1414 25.175 0.6714 3.1472 1.1311 2.2043
Inf 1.7414 1.6774 2.1554 1.7028 2.0792 1.4943 1.1016

N 1/R L1 C2 L3 C4 L5 C6 L7 N 1/R L1 C2 L3 C4 L5 C6 L7

Note: (ω = 1 rad/s for Atn = 3dB)


Microwave Devices, Circuits and Subsystems for Communications Engineering
Table 6.8 Low pass prototype filter: Bessel

N R C1 L2 C3 L4 C5 L6 C7 N R C1 L2 C3 L4 C5 L6 C7
Filters

2 1.000 0.5755 2.1478 5 1.000 0.1743 0.5072 0.8040 1.1110 2.2582


1.111 0.5084 2.3097 0.900 0.1926 0.4542 0.8894 0.9945 2.4328
1.250 0.4433 2.5096 0.800 0.2154 0.4016 0.9959 0.8789 2.6497
1.429 0.3801 2.7638 0.700 0.2447 0.3494 1.1323 0.7642 2.9272
1.667 0.3191 3.0993 0.600 0.2836 0.2977 1.3138 0.6506 3.2952
2.000 0.2601 3.5649 0.500 0.3380 0.2465 1.5672 0.5382 3.8077
2.500 0.2032 4.2577 0.400 0.4194 0.1958 1.9464 0.4270 4.5731
3.333 0.1486 5.4050 0.300 0.5548 0.1457 2.5768 0.3174 5.8433
5.000 0.0965 7.6876 0.200 0.8251 0.0964 3.8352 0.2095 8.3747
10.00 0.0469 14.510 0.100 1.6349 0.0478 7.6043 0.1036 15.949
Inf 1.3617 0.4539 Inf 1.5125 1.0232 0.7531 0.4729 0.1618
3 1.000 0.3374 0.9705 2.2034 6 1.000 0.1365 0.4002 0.6392 0.8538 1.1126 2.2645
0.900 0.3708 0.8650 2.3745 1.111 0.1223 0.4429 0.5732 0.9456 0.9964 2.4388
0.800 0.4124 0.7609 2.5867 1.250 0.1082 0.4961 0.5076 1.0600 0.8810 2.6554
0.700 0.4657 0.6584 2.8575 1.429 0.0943 0.5644 0.4424 1.2069 0.7665 2.9325
0.600 0.5365 0.5576 3.2159 1.667 0.0804 0.6553 0.3775 1.4022 0.6530 3.3001
0.500 0.6353 0.4587 3.7144 2.000 0.0666 0.7824 0.3131 1.6752 0.5405 3.8122
0.400 0.7829 0.3618 4.4573 2.500 0.0530 0.9725 0.2492 2.0837 0.4292 4.5770
0.300 1.0283 0.2673 6.6888 3.333 0.0395 1.2890 0.1859 2.7633 0.3193 5.8467
0.200 1.5171 0.1752 8.1403 5.000 0.0261 1.9209 0.1232 4.1204 0.2110 8.3775
0.100 2.9825 0.0860 15.470 10.00 0.0130 3.8146 0.0612 8.1860 0.1045 15.951
Inf 1.4631 0.8427 0.2926 Inf 1.5124 1.0329 0.8125 0.6072 0.3785 0.1287
4 1.000 0.2334 0.6725 1.0815 2.2404 7 1.000 0.1106 0.3259 0.5249 0.7020 0.8690 1.1052 2.2659
1.111 0.2085 0.7423 0.9670 2.4143 0.900 0.1224 0.2923 0.5815 0.6302 0.9630 0.9899 2.4396
1.250 0.1839 0.8292 0.8534 2.6304 0.800 0.1372 0.2589 0.6521 0.5586 1.0803 0.8754 2.6556
1.429 0.1596 0.9406 0.7410 2.9066 0.700 0.1562 0.2257 0.7428 0.4873 1.2308 0.7618 2.9319
1.667 0.1356 1.0886 0.6299 3.2727 0.600 0.1815 0.1927 0.8634 0.4163 1.4312 0.6491 3.2984
2.000 0.1120 1.2952 0.5202 3.7824 0.500 0.2168 0.1599 1.0321 0.3457 1.7111 0.5374 3.8090
2.500 0.0887 1.6040 0.4120 4.5430 0.400 0.2698 0.1274 1.2847 0.2755 2.1304 0.4269 4.5718
3.333 0.0658 2.1174 0.3056 5.8048 0.300 0.3579 0.0951 1.7051 0.2058 2.8280 0.3177 5.8380
5.000 0.0434 3.1416 0.2013 8.3185 0.200 0.5338 0.0630 2.5448 0.1365 4.2214 0.2100 8.3623
10.00 0.0214 6.2086 0.0993 15.837 0.100 1.0612 0.0313 5.0616 0.0679 8.3967 0.1040 15.917
Inf 1.5012 0.9781 0.6127 0.2114 Inf 1.5087 1.0293 0.8345 0.6752 0.5031 0.3113 0.1054

N 1/R L1 C2 L3 C4 L5 C6 L7 N 1/R L1 C2 L3 C4 L5 C6 L7
447

Note: (ω = 1 rad/s for Atn = 3dB)


448 Microwave Devices, Circuits and Subsystems for Communications Engineering

Table 6.9 Low pass prototype filter: Elliptic

θ ωs Amin Ω2 C1 C2 L2 C3

1 57.2987 103.56 66.1616 0.6395 0.0002 0.9786 0.6395


2 28.6537 85.50 33.0839 0.6390 0.0009 0.9776 0.6390
3 19.1073 74.93 22.0595 0.6381 0.0021 0.9761 0.6381
4 14.3356 67.43 16.5483 0.6370 0.0037 0.9739 0.6370
5 11.4737 61.61 13.2424 0.6354 0.0059 0.9711 0.6354
6 9.5668 56.85 11.0392 0.6336 0.0085 0.9676 0.6336
7 8.2055 52.82 9.4661 0.6314 0.0116 0.9636 0.6314
8 7.1853 49.33 8.2868 0.6289 10.0152 0.9589 0.6289
9 6.3925 46.25 7.3700 0.6261 0.0193 0.9536 0.6261
10 5.7588 43.49 6.6370 0.6229 0.0240 0.9477 0.6229
11 5.2408 41.00 6.0377 0.6194 0.0291 0.9411 0.6194
12 4.8097 38.71 5.5386 0.6155 0.0349 0.9339 0.6155
13 4.4454 36.61 5.1166 0.6113 0.0412 0.9261 0.6113
14 4.1336 34.66 4.7552 0.6068 0.0482 0.9177 0.6068
15 3.8637 32.85 4.4423 0.6020 0.0558 0.9087 0.6020
16 3.6280 31.14 4.1688 0.5968 0.0640 0.8991 0.5968
17 3.4203 29.54 3.9277 0.5913 0.0729 0.8888 0.5913
18 3.2361 28.03 3.7137 0.5855 0.0826 0.8780 0.5855
19 3.0716 26.60 3.5224 0.5793 0.0930 0.8665 0.5793
20 2.9238 25.24 3.3505 0.5728 0.1043 0.8545 0.5728
21 2.7904 23.95 3.1951 0.5661 0.1164 0.8418 0.5661
22 2.6695 22.71 3.0541 0.5590 0.1294 0.8286 0.5590
23 2.5593 21.53 2.9256 0.5515 0.1434 0.8148 0.5515
24 2.4586 20.40 2.8079 0.5438 0.1585 0.8004 0.5438
25 2.3662 19.31 2.6999 0.5358 0.1747 0.7855 0.5358
26 2.2812 18.27 2.6003 0.5275 0.1921 0.7700 0.5275
27 2.2027 17.26 2.5083 0.5189 0.2108 0.7540 0.5189
28 2.1301 16.30 2.4231 0.5100 0.2309 0.7375 0.5100
29 2.0627 15.37 2.3438 0.5009 0.2526 0.7205 0.5009
30 2.0000 14.47 2.2701 0.4915 0.2760 0.7031 0.4915
31 1.9416 13.61 2.2012 0.4819 0.3012 0.6852 0.4819
32 1.8871 12.77 2.1368 0.4720 0.3284 0.6669 0.4720

θ ωs Amin Ω2 L1 L2 C2 L3

Note: (N = 3, R = 1, RdB = 0.01)


Filters 449

Table 6.10 Low pass prototype filter: Elliptic

θ ωs Amin A2 C1 C2 L2 C3

1 57.2987 115.77 66.1616 1.1893 0.0002 1.1540 1.1893


2 28.6537 97.70 33.0839 1.1889 0.0008 1.1533 1.1889
3 19.1073 87.13 22.0595 1.1881 0.0018 1.1522 1.1881
4 14.3356 79.63 16.5483 1.1870 0.0032 1.1507 1.1870
5 11.4737 73.81 13.2424 1.1856 0.0050 1.1488 1.1856
6 9.5668 69.05 11.0392 1.1839 0.0072 1.1464 1.1839
7 8.2055 65.03 9.4661 1.1819 0.0098 1.1436 1.1819
8 7.1853 61.54 8.2868 1.1796 0.0128 1.1404 1.1796
9 6.3925 58.46 7.3700 1.1770 0.0162 1.1367 1.1770
10 5.7588 55.70 6.6370 1.1740 0.0200 1.1326 1.1740
11 5.2408 53.20 6.0377 1.1708 0.0243 1.1281 1.1708
12 4.8097 50.92 5.5386 1.1672 0.0290 1.1231 1.1672
13 4.4454 48.82 5.1166 1.1634 0.0342 1.1177 1.1634
14 4.1336 46.87 4.7552 1.1592 0.0398 1.1119 1.1592
15 3.8637 45.05 4.4423 1.1547 0.0458 1.1057 1.1547
16 3.6280 43.35 4.1688 1.1500 0.0524 1.0990 1.1500
17 3.4203 41.75 3.9277 1.1449 0.0594 1.0919 1.1449
18 3.2361 40.23 3.7137 1.1395 0.0669 1.0844 1.1395
19 3.0716 38.80 3.5224 1.1338 0.0749 1.0764 1.1338
20 2.9238 37.44 3.3505 1.1278 0.0834 1.0681 1.1278
21 2.7904 36.14 3.1951 1.1215 0.0925 1.0593 1.1215
22 2.6695 34.90 3.0541 1.1149 0.1021 1.0500 1.1149
23 2.5593 33.71 2.9256 1.1080 0.1123 1.0404 1.1080
24 2.4586 32.57 2.8079 1.1008 0.1231 1.0303 1.1008
25 2.3662 31.47 2.6999 1.0933 0.1345 1.0199 1.0933
26 2.2812 30.41 2.6003 1.0855 0.1466 1.0090 1.0855
27 2.2027 29.39 2.5083 1.0773 0.1593 0.9976 1.0773
28 2.1301 28.41 2.4231 1.0689 0.1728 0.9859 1.0689
29 2.0627 27.45 2.3438 1.0602 0.1869 0.9738 1.0602
30 2.0000 26.53 2.2701 1.0512 0.2019 0.9612 1.0512
31 1.9416 25.63 2.2012 1.0420 0.2176 0.9483 1.0420
32 1.8871 24.76 2.1368 1.0324 0.2343 0.9349 1.0324
33 1.8361 23.92 2.0765 1.0225 0.2518 0.9212 1.0225
34 1.7883 23.09 2.0199 1.0123 0.2702 0.9070 1.0123
35 1.7434 22.29 1.9666 1.0019 0.2897 0.8925 1.0019
36 1.7013 21.51 1.9165 0.9912 0.3103 0.8776 0.9912
37 1.6616 20.74 1.8692 0.9802 0.3320 0.8623 0.9802
38 1.6243 20.00 1.8245 0.9689 0.3549 0.8466 0.9689
39 1.5890 19.27 1.7823 0.9573 0.3791 0.8305 0.9573
40 1.5557 18.56 1.7423 0.9455 0.4047 0.8141 0.9455
41 1.5243 17.86 1.7044 0.9334 0.4318 0.7973 0.9334
42 1.4945 17.18 1.6684 0.9210 0.4605 0.7801 0.9210
43 1.4663 16.52 1.6343 0.9084 0.4909 0.7627 0.9084
44 1.4396 15.86 1.6018 0.8955 0.5232 0.7448 0.8955
45 1.4142 15.22 1.5710 0.8823 0.5576 0.7267 0.8823
46 1.3902 14.60 1.5415 0.8689 0.5942 0.7082 0.8689
47 1.3673 13.98 1.5135 0.8553 0.6331 0.6895 0.8553
48 1.3456 13.38 1.4868 0.8415 0.6747 0.6705 0.8415
49 1.3250 12.79 1.4613 0.8274 0.7192 0.6511 0.8274
50 1.3054 12.22 1.4369 0.8131 0.7668 0.6316 0.8131

θ ωs Amin A2 L1 L2 C2 L3

Note: (N = 3, R = 1, RdB = 0.18)


450 Microwave Devices, Circuits and Subsystems for Communications Engineering

Table 6.11 Low pass prototype filter: Elliptic

θ ωs Amin A2 C1 C2 L2 C3 L4

6 10.350843 88.5 11.367741 0.7174 0.006461 1.198 1.330 0.6549


7 8.876727 83.1 9.747389 0.7154 0.008813 1.194 1.329 0.6552
8 7.771760 78.5 8.532615 0.7130 0.01154 1.190 1.327 0.6555
9 6.912894 74.4 7.588226 0.7103 0.01464 1.186 1.325 0.6558
10 6.226301 70.7 6.833109 0.7073 0.01814 1.181 1.323 0.6561
11 5.664999 67.3 6.215646 0.7040 0.02202 1.176 1.321 0.6565
12 5.197666 64.3 5.701423 0.7003 0.02630 1.170 1.318 0.6569
13 4.802620 61.5 5.266618 0.6963 0.03100 1.163 1.316 0.6574
14 4.464371 58.9 4.894214 0.6920 0.03612 1.156 1.313 0.6579
15 4.171563 56.5 4.571732 0.6874 0.04166 1.148 1.310 0.6584
16 3.915678 54.2 4.289813 0.6824 0.04766 1.140 1.306 0.6590
17 3.690200 52.1 4.041300 0.6771 0.05411 1.132 1.303 0.6596
18 3.490065 50.1 3.820626 0.6715 0.06103 1.122 1.299 0.6603
19 3.311272 48.1 3.623399 0.6655 0.06845 1.113 1.295 0.6610
20 3.150622 46.3 3.446101 0.6592 0.07637 1.103 1.291 0.6617
21 3.005526 44.6 3.285888 0.6526 0.08482 1.092 1.286 0.6624
22 2.873864 42.9 3.140431 0.6456 0.09383 1.081 1.282 0.6632
23 2.753885 41.3 3.007807 0.6383 0.1034 1.069 1.277 0.6641
24 2.644133 39.8 2.886413 0.6306 0.1136 1.057 1.272 0.6649
25 2.543380 38.4 2.774903 0.6226 0.1244 1.044 1.267 0.6658
26 2.450592 37.0 2.672139 0.6143 0.1359 1.030 1.262 0.6668
27 2.364885 35.6 2.577149 0.6055 0.1481 1.017 1.256 0.6677
28 2.285502 34.3 2.489103 0.5964 0.1611 1.002 1.250 0.6687
29 2.211792 33.0 2.407283 0.5870 0.1748 0.9872 1.244 0.6698
30 2.143189 31.8 2.331070 0.5772 0.1894 0.9717 1.238 0.6708
31 2.079202 30.6 2.259921 0.5670 0.2049 0.9558 1.232 0.6719
32 2.019399 29.4 2.193363 0.5564 0.2213 0.9393 1.226 0.6730
33 1.963403 28.3 2.130982 0.5455 0.2388 0.9223 1.219 0.6742
34 1.910879 27.2 2.072410 0.5341 0.2573 0.9048 1.212 0.6753
35 1.861534 26.1 2.017322 0.5224 0.2771 0.8868 1.205 0.6765
36 1.815103 25.1 1.965429 0.5103 0.2982 0.8683 1.198 0.6777
37 1.771354 24.0 1.916475 0.4978 0.3206 0.8492 1.191 0.6789
38 1.730076 23.0 1.870229 0.4848 0.3446 0.8297 1.184 0.6801
39 1.691083 22.1 1.826485 0.4715 0.3702 0.8098 1.177 0.6813
40 1.654204 21.1 1.785057 0.4577 0.3976 0.7893 1.169 0.6825

θ ωs Amin A2 L1 L2 C2 L3 C4

Note: (N = 4, R = 1.1, RdB = 0.01)


Filters 451

Table 6.12 Low pass prototype filter: Elliptic

θ ωs Amin A2 C1 C2 L2 C3 L4

6 10.35084 100.7 11.36774 1.260 0.006028 1.284 1.932 0.8431


7 8.876727 95.3 9.747390 1.258 0.008216 1.281 1.930 0.843
8 7.771760 90.7 8.532615 1.255 0.01074 1.278 1.928 0.8440
9 6.912894 86.6 7.588226 1.253 0.01362 1.275 1.926 0.8442
10 6.226301 82.9 6.833109 1.250 0.01685 1.271 1.924 0.8443
11 5.664999 79.6 6.215646 1.247 0.02043 1.267 1.921 0.8445
12 5.197666 76.5 5.701423 1.243 0.02436 1.263 1.918 0.8448
13 4.802620 73.7 5.266618 1.239 0.02866 1.258 1.915 0.8450
14 4.464371 71.1 4.894214 1.235 0.03333 1.253 1.912 0.8453
15 4.171563 68.7 4.571731 1.231 0.03837 1.247 1.908 0.8456
16 3.915678 66.4 4.289813 1.226 0.04380 1.241 1.904 0.8459
17 3.690200 64.3 4.041300 1.221 0.04961 1.234 1.900 0.8462
18 3.490065 62.3 3.820626 1.216 0.05581 1.227 1.895 0.8465
19 3.311272 60.3 3.623399 1.210 0.06242 1.220 1.891 0.8469
20 3.150622 58.5 3.446101 1.204 0.06944 1.213 1.886 0.8473
21 3.005526 56.8 3.285888 1.198 0.07689 1.205 1.881 0.8477
22 2.873864 55.1 3.140431 1.191 0.08476 1.196 1.875 0.8481
23 2.753885 53.6 3.007807 1.184 0.09309 1.187 1.870 0.8485
24 2.644133 52.0 2.886413 1.177 0.1019 1.178 1.864 0.8490
25 2.543380 50.6 2.774903 1.169 0.1111 1.169 1.858 0.8494
26 2.450592 49.2 2.672139 1.161 0.1209 1.159 1.851 0.8499
27 2.364885 47.8 2.577149 1.153 0.1311 1.148 1.845 0.8505
28 2.285502 46.5 2.489103 1.145 0.1419 1.138 1.838 0.8510
29 2.211792 45.2 2.407283 1.136 0.1532 1.126 1.831 0.8516
30 2.143189 44.0 2.331070 1.127 0.1651 1.115 1.824 0.8521
31 2.079202 42.8 2.259921 1.117 0.1775 1.103 1.816 0.8527
32 2.019399 41.6 2.193363 1.108 0.1906 1.091 1.808 0.8533
33 1.963403 40.5 2.130982 1.097 0.2043 1.078 1.800 0.8540
34 1.910879 39.4 2.072410 1.087 0.2186 1.065 1.792 0.8546
35 1.861534 38.3 2.017322 1.076 0.2337 1.051 1.784 0.8553
36 1.815103 37.2 1.965429 1.065 0.2495 1.038 1.775 0.8560
37 1.771354 36.2 1.916475 1.054 0.2661 1.023 1.766 0.8567
38 1.730076 35.2 1.870229 1.042 0.2835 1.009 1.757 0.8574
39 1.691083 34.2 1.826485 1.030 0.3017 0.9936 1.748 0.8581
40 1.654204 33.3 1.785057 1.017 0.3208 0.9782 1.738 0.8589

θ ωs Amin A2 L1 L2 C2 L3 C4

Note: (N = 4, R = 1.5, RdB = 0.18)


452

Table 6.13 Low pass prototype filter: Elliptic

θ ωs Amin A2 A4 C1 C2 L2 C3 C4 L4 C5

2 28.6537 167.86 48.7389 30.1274 0.7661 0.0003 1.3099 1.5877 0.0008 1.3091 0.7656
3 19.1073 150.25 32.4927 20.0893 0.7658 0.0007 1.3095 1.5868 0.0018 1.3075 0.7646
4 14.3356 137.74 24.3697 15.0716 0.7654 0.0012 1.3088 1.5855 0.0033 1.3054 0.7633
5 11.4737 128.04 19.4959 12.0620 0.7648 0.0020 1.3080 1.5839 0.0052 1.3026 0.7615
6 9.5668 120.11 16.2468 10.0565 0.7641 0.0029 1.3070 1.5820 0.0076 1.2993 0.7594
7 8.2055 113.40 13.9260 8.6247 0.7632 0.0039 1.3058 1.5796 0.0103 1.2953 0.7569
8 7.1853 107.59 12.1854 7.5516 0.7623 0.0051 1.3044 1.5770 0.0135 1.2907 0.7540
9 6.3925 102.45 10.8316 6.7175 0.7612 0.0065 1.3028 1.5739 0.0172 1.2855 0.7507
10 5.7588 97.86 9.7486 6.0507 0.7600 0.0080 1.3011 1.5706 0.0213 1.2797 0.7470
11 5.2408 93.69 8.8625 5.5057 0.7586 0.0098 1.2991 1.5669 0.0259 1.2733 0.7429
12 4.8097 89.89 8.1241 5.0520 0.7572 0.0116 1.2970 1.5628 0.0309 1.2663 0.7384
13 4.4454 86.39 7.4993 4.6684 0.7556 0.0137 1.2947 1.5584 0.0364 1.2586 0.7335
14 4.1336 83.14 6.9638 4.3401 0.7538 0.0159 1.2922 1.5536 0.0424 1.2504 0.7283
15 3.8637 80.11 6.4997 4.0559 0.7519 0.0183 1.2895 1.5485 0.0489 1.2416 0.7226
16 3.6280 77.27 6.0936 3.8076 0.7499 0.0209 1.2866 1.5431 0.0559 1.2321 0.7165
17 3.4203 74.60 5.7353 3.5888 0.7478 0.0236 1.2836 1.5374 0.0635 1.2221 0.7101
18 3.2361 72.08 5.4168 3.3946 0.7455 0.0266 1.2803 1.5313 0.0716 1.2115 0.7032
19 3.0716 69.69 5.1318 3.2212 0.7431 0.0297 1.2768 1.5249 0.0802 1.2002 0.6959
20 2.9238 67.41 4.8753 3.0654 0.7406 0.0330 1.2732 1.5182 0.0895 1.1884 0.6883
21 2.7904 65.25 4.6433 2.9246 0.7379 0.0365 1.2694 1.5112 0.0994 1.1760 0.6802
22 2.6695 63.18 4.4323 2.7970 0.7350 0.0402 1.2653 1.5038 0.1099 1.1630 0.6717
23 2.5593 61.20 4.2397 2.6807 0.7321 0.0441 1.2611 1.4962 0.1210 1.1494 0.6628
24 2.4586 59.29 4.0631 2.5743 0.7290 0.0482 1.2567 1.4882 0.1329 1.1353 0.6534
25 2.3662 57.46 3.9007 2.4767 0.7257 0.0524 1.2520 1.4800 0.1454 1.1205 0.6437
26 2.2812 55.70 3.7507 2.3868 0.7223 0.0569 1.2472 1.4715 0.1588 1.1052 0.6335
27 2.2027 54.00 3.6119 2.3038 0.7187 0.0617 1.2421 1.4627 0.1729 1.0893 0.6229
28 2.1301 52.35 3.4829 2.2270 0.7150 0.0666 1.2369 1.4537 0.1879 1.0729 0.6118
Microwave Devices, Circuits and Subsystems for Communications Engineering

29 2.0627 50.76 3.3629 2.1556 0.7112 0.0718 1.2314 1.4444 0.2038 1.0559 0.6003
30 2.0000 49.22 3.2508 2.0892 0.7072 0.0772 1.2257 1.4348 0.2206 1.0383 0.5884
31 1.9416 47.72 3.1460 2.0274 0.7030 0.08281 1.2198 1.4250 0.2384 1.0201 0.5760
Filters

32 1.8871 46.27 3.0476 1.9695 0.6987 0.08871 1.2136 1.4150 0.2574 1.0015 0.5631
33 1.8361 44.85 2.9553 1.9154 0.6942 0.09481 1.2073 1.4048 0.2774 0.9822 0.5498
34 1.7883 43.47 2.8683 1.8646 0.6896 0.10121 1.2007 1.3943 0.2988 0.9625 0.5360
35 1.7434 42.13 2.7864 1.8170 0.6847 0.1078 1.1938 1.3837 0.3214 0.9422 0.5217
36 1.7013 40.81 2.7089 1.7722 0.6798 0.1148 1.1867 1.3729 0.3455 0.9214 0.5070
37 1.6616 39.53 2.6356 1.7299 0.6746 0.1220 1.1794 1.3619 0.3712 0.9001 0.4917
38 1.6243 38.28 2.5662 1.6901 0.6693 0.1295 1.1717 1.3508 0.3985 0.8782 0.4759
39 1.5890 37.05 2.5003 1.6525 0.6637 0.1374 1.1638 1.3396 0.4278 0.8559 0.4595
40 1.5557 35.85 2.4377 1.6170 0.6580 0.1456 1.1556 1.3282 0.4590 0.8331 0.4426
41 1.5243 34.67 2.3781 1.5833 0.6521 0.1541 1.1472 1.3168 0.4924 0.8099 0.4252
42 1.4945 33.52 2.3213 1.5515 0.6460 0.1630 1.1384 1.3053 0.5283 0.7862 0.4072
43 1.4663 32.38 2.2672 1.5213 0.6397 0.1722 1.1292 1.2937 0.5669 0.7620 0.3885
44 1.4396 31.27 2.2154 1.4926 0.6332 0.1819 1.1198 1.2822 0.6085 0.7375 0.3693
45 1.4142 30.17 2.1660 1.4654 0.6265 0.1920 1.1099 1.2706 0.6535 0.7125 0.3494
46 1.3902 29.09 2.1187 1.4396 0.6195 0.2025 1.0997 1.2591 0.7022 0.6871 0.3288
47 1.3673 28.03 2.0733 1.4150 0.6124 0.2135 1.0891 1.2478 0.7550 0.6614 0.3075
48 1.3456 26.99 2.0299 1.3916 0.6050 0.2251 1.0780 1.2365 0.8126 0.6354 0.2855
49 1.3250 25.95 1.9881 1.3693 0.5973 0.2372 1.0665 1.2254 0.8756 0.6090 0.2628
50 1.3054 24.94 1.9480 1.3481 0.5894 0.2498 1.0545 1.2145 0.9446 0.5824 0.2392
51 1.2868 23.93 1.9095 1.3279 0.5813 0.2632 1.0420 1.2039 1.0206 0.5556 0.2147
52 1.2690 22.94 1.8724 1.3087 0.5729 0.2772 1.0289 1.1937 1.1047 0.5285 0.1894
53 1.2521 21.96 1.8366 1.2903 0.5642 0.2920 1.0152 1.1838 1.1980 0.5013 0.1631
54 1.2361 20.99 1.8021 1.2728 0.5552 0.3076 1.0008 1.1744 1.3020 0.4740 0.1357
55 1.2208 20.04 1.7689 1.2561 0.5460 0.3242 0.9858 1.1656 1.4187 0.4467 0.1073
56 1.2062 19.09 1.7368 1.2402 0.5364 0.3417 0.9700 1.1575 1.5502 0.4194 0.0777
57 1.1924 18.15 1.7057 1.2250 0.5265 0.3605 0.9533 1.1501 1.6993 0.3921 0.0468
58 1.1792 17.23 1.6757 1.2104 0.5163 0.3805 0.9358 1.1437 1.8693 0.3651 0.0145

θ ωs Amin A2 A4 L1 L2 C2 L3 L4 C4 L5

Note: (N = 5, R = 1, RdB = 0.01)


453
454

Table 6.14 Low pass prototype filter: Elliptic

θ ωs Amin A2 A4 C1 C2 L2 C3 C4 L4 C5

2 28.6537 180.07 48.7389 30.1274 1.30163 0.00031 1.34523 2.12770 0.00082 1.34459 1.30112
3 19.1073 162.45 32.4927 20.0893 1.30130 0.00070 1.34483 2.12660 0.00184 1.34339 1.30016
4 14.3356 149.95 24.3697 15.0716 1.30084 0.00125 1.34426 2.12507 0.00328 1.34170 1.29881
5 11.4737 140.25 19.4959 12.0620 1.30024 0.00196 1.34353 2.12311 0.00513 1.33954 1.29708
6 9.5668 132.32 16.2468 10.0565 1.29951 0.00282 1.34264 2.12070 0.00740 1.33689 1.29496
7 8.2055 125.61 13.9260 8.6247 1.29865 0.00384 1.34159 2.11786 0.01008 1.33376 1.29246
8 7.1853 119.80 12.1854 7.5516 1.29766 0.00502 1.34037 2.11459 0.01318 1.33015 1.28957
9 6.3925 114.66 10.8316 6.7175 1.29653 0.00637 1.33899 2.11088 0.01671 1.32607 1.28630
10 5.7588 110.06 9.7486 6.0507 1.29527 0.00787 1.33744 2.10675 0.02067 1.32150 1.28264
11 5.2408 105.90 8.8625 5.5057 1.29387 0.00953 1.33573 2.10217 0.02506 1.31646 1.27859
12 4.8097 102.10 8.1241 5.0520 1.29234 0.01136 1.33386 2.09717 0.02989 1.31094 1.27417
13 4.4454 98.59 7.4993 4.6684 1.29067 0.01335 1.33182 2.09172 0.03516 1.30495 1.26936
14 4.1336 95.34 6.9638 4.3401 1.28887 0.01551 1.32961 2.08588 0.04089 1.29848 1.26416
15 3.8637 92.32 6.4997 4.0559 1.28693 0.01783 1.32724 2.07959 0.04707 1.29154 1.25858
16 3.6280 89.48 6.0936 3.8076 1.28485 0.02033 1.32470 2.07288 0.05371 1.28413 1.25261
17 3.4203 86.81 5.7353 3.5888 1.28263 0.02300 1.32199 2.06574 0.06084 1.27625 1.24627
18 3.2361 84.29 5.4168 3.3946 1.28027 0.02584 1.31911 2.05819 0.06844 1.26790 1.23953
19 3.0716 81.89 5.1318 3.2212 1.27778 0.02885 1.31607 2.05021 0.07655 1.25909 1.23241
20 2.9238 79.62 4.8753 3.0654 1.27514 0.03205 1.31285 2.04182 0.08515 1.24981 1.22491
21 2.7904 77.46 4.6433 2.9246 1.27236 0.03542 1.30945 2.03301 0.09428 1.24007 1.21703
22 2.6695 75.39 4.4323 2.7970 1.26943 0.03898 1.30589 2.02379 0.10393 1.22987 1.20876
23 2.5593 73.40 4.2397 2.6807 1.26636 0.04272 1.30215 2.01416 0.11414 1.21921 1.20010
24 2.4586 71.50 4.0631 2.5743 1.26314 0.04666 1.29823 2.00412 0.12490 1.20809 1.19107
25 2.3662 69.67 3.9007 2.4767 1.25978 0.05079 1.29413 1.99368 0.13625 1.19652 1.18164
26 2.2812 67.91 3.7507 2.3868 1.25262 0.05511 1.28985 1.98283 0.14819 1.18450 1.17183
27 2.2027 66.21 3.6119 2.3038 1.25259 0.05963 1.28540 1.97159 0.16075 1.17203 1.16164
28 2.1301 64.56 3.4829 2.2270 1.24877 0.06436 1.28075 1.95995 0.17396 1.15911 1.15106
29 2.0627 62.97 3.3629 2.1556 1.22480 0.06930 1.27592 1.94792 0.18783 1.14576 1.14010
Microwave Devices, Circuits and Subsystems for Communications Engineering

30 2.0000 61.43 3.2508 2.0892 1.24067 0.07446 1.27091 1.93550 0.20239 1.13196 1.12874
31 1.9416 59.93 3.1460 2.0274 1.23638 0.07983 1.26570 1.92270 0.21768 1.11772 1.11700
32 1.8871 58.47 3.0476 1.9695 1.23192 0.08543 1.26030 1.90952 0.23371 1.10305 1.10487
Filters

33 1.8361 57.06 2.9553 1.9154 1.22731 0.09126 1.25470 1.89596 0.25054 1.08795 1.09235
34 1.7883 55.68 2.8683 1.8846 1.22252 0.09732 1.24890 1.88203 0.26819 1.07242 1.07944
35 1.7434 54.33 2.7864 1.8170 1.21757 0.10363 1.24290 1.86773 0.28671 1.05648 1.06614
36 1.7013 53.02 2.7089 1.7722 1.21244 0.11019 1.23669 1.85307 0.30614 1.04011 1.05244
37 1.6616 51.74 2.6356 1.7299 1.20714 0.11701 1.23028 1.83806 0.32654 1.02332 1.03835
38 1.6243 50.49 2.5662 1.6901 1.20166 0.12410 1.22364 1.82269 0.34795 1.00613 1.02386
39 1.5890 49.26 2.5003 1.6525 1.19600 0.13146 1.21679 1.80698 0.37044 0.98853 1.00897
40 1.5557 48.06 2.4377 1.6170 1.19015 0.13911 1.20971 1.79093 0.39408 0.97053 0.99368
41 1.5243 46.88 2.3781 1.5833 1.18411 0.14706 1.20241 1.77455 0.41894 0.95213 0.97798
42 1.4945 45.72 2.3213 1.5515 1.17787 0.15532 1.19486 1.75784 0.44510 0.93335 0.96187
43 1.4663 44.59 2.2672 1.5213 1.17144 0.16389 1.18708 1.74081 0.47265 0.91417 0.94535
44 1.4396 43.47 2.2154 1.4926 1.16480 0.17280 1.17904 1.72347 0.50170 0.89462 0.92841
45 1.4142 42.38 2.1660 1.4654 1.15794 0.18206 1.17075 1.70583 0.53236 0.87470 0.91105
46 1.3902 41.30 2.1187 1.4396 1.15088 0.19169 1.16219 1.68789 0.56476 0.85441 0.89326
47 1.3673 40.23 2.0733 1.4150 1.14359 0.20169 1.15336 1.66967 0.59903 0.83376 0.87504
48 1.3456 39.19 2.0299 1.3916 1.13607 0.21210 1.14425 1.65117 0.63534 0.81276 0.85638
49 1.3250 38.15 1.9881 1.3693 1.12831 0.22293 1.13484 1.63241 0.67386 0.79141 0.83727
50 1.3054 37.13 1.9480 1.3481 1.12031 0.23421 1.12513 1.61339 0.71481 0.76973 0.81771
51 1.2868 36.12 1.9095 1.3279 1.11206 0.24596 1.11509 1.59413 0.75841 0.74773 0.79768
52 1.2690 35.13 1.8724 1.3087 1.10354 0.25821 1.10473 1.57465 0.80492 0.72541 0.77717
53 1.2521 34.14 1.8366 1.2903 1.09476 0.27099 1.09401 1.55494 0.85465 0.70278 0.75619
54 1.2361 33.17 1.0821 1.2728 1.08569 0.28433 1.08293 1.53504 0.90794 0.67986 0.73470
55 1.2208 32.20 1.7689 1.2561 1.07633 0.29828 1.07147 1.51496 0.96518 0.65667 0.71270
56 1.2062 31.25 1.7368 1.2402 1.06666 0.31288 1.05960 1.49471 1.02684 0.63320 0.69016
57 1.1924 30.30 1.7057 1.2250 1.05668 0.32817 1.04731 1.47431 1.09344 0.60949 0.66709
58 1.1792 29.36 1.6757 1.2104 1.04636 0.34422 1.03456 1.45379 1.16561 0.58554 0.64344
59 1.1666 28.42 1.6467 1.1966 1.03570 0.36109 1.02134 1.43317 1.24407 0.56138 0.61920
60 1.1547 27.49 1.6185 1.1834 1.02467 0.37885 1.00760 1.41247 1.32969 0.53702 0.59435

θ ωs Amin A2 A4 L1 L2 C2 L3 L4 C4 L5
455

Note: (N = 5, R = 1, RdB = 0.18)


Table 6.15 Low pass prototype filter: Elliptic
456

θ ωs Amin A2 A4 C1 C2 L2 C3 C4 L4 C5 L6

16 3.751039 112.5 5.452491 3.888329 1.299 0.0250 1.344 2.142 0.0468 1.412 2.017 0.8828
17 3.535748 109.3 5.133037 3.664543 1.296 0.0283 1.341 2.135 0.0530 1.405 2.012 0.8830
18 3.344698 106.3 4.849152 3.465915 1.293 0.0318 1.337 2.126 0.0596 1.397 2.006 0.8831
19 3.174064 103.4 4.595218 3.288476 1.290 0.0355 1.333 2.118 0.0666 1.389 2.000 0.8833
20 3.020785 100.7 4.366743 3.129050 1.286 0.0395 1.328 2.108 0.0740 1.380 1.993 0.8835
21 2.882384 98.1 4.160091 2.985065 1.283 0.0436 1.324 2.099 0.0818 1.371 1.987 0.8837
22 2.756834 95.6 3.972284 2.854418 1.279 0.0480 1.319 2.089 0.0901 1.362 1.979 0.8839
23 2.642462 93.3 3.800865 2.735370 1.275 0.0527 1.314 2.078 1.0989 1.352 1.972 0.8841
24 2.537873 91.0 3.643786 2.626475 1.270 0.0576 1.309 2.067 0.1081 1.341 1.964 0.8843
25 2.441895 88.8 3.499325 2.526516 1.266 0.0627 1.303 2.055 0.1177 1.331 1.956 0.8845
26 2.353536 86.7 3.366027 2.434463 1.261 0.0680 1.297 2.043 0.1279 1.320 1.948 0.8848
27 2.271953 84.6 3.242651 2.349441 1.256 0.0736 1.291 2.031 0.1385 1.308 1.939 0.8850
28 2.196422 82.6 3.128134 2.270699 1.251 0.0795 1.285 2.018 0.1497 1.296 1.930 0.8853
29 2.126320 80.7 3.021559 2.197588 1.246 0.0857 1.279 2.005 0.1613 1.284 1.921 0.8855
30 2.061105 78.9 2.922132 2.129549 1.240 0.0921 1.272 1.991 0.1735 1.271 1.911 0.8858
31 2.000308 77.1 2.829162 2.066092 1.235 0.0988 1.265 1.977 0.1863 1.257 1.901 0.8861
32 1.943517 75.3 2.742042 2.006790 1.229 0.1057 1.258 1.962 0.1996 1.244 1.891 0.8864
33 1.890370 73.6 2.660241 1.951268 1.223 0.1130 1.250 1.947 0.2136 1.230 1.881 0.8867
34 1.840548 72.0 2.583290 1.899195 1.216 0.1206 1.243 1.931 0.2281 1.215 1.870 0.8870
35 1.793769 70.4 2.510772 1.850277 1.210 0.1285 1.235 1.915 0.2433 1.200 1.859 0.8873
36 1.749781 68.8 2.442318 1.804254 1.203 0.1367 1.226 1.899 0.2592 1.185 1.847 0.8877
37 1.708362 67.3 2.377598 1.760893 1.196 0.1452 1.218 1.882 0.2758 1.169 1.835 0.8880
38 1.669312 65.8 2.316318 1.719987 1.189 0.1541 1.209 1.864 0.2931 1.153 1.823 0.8884
39 1.632449 64.3 2.258212 1.681350 1.181 0.1634 1.200 1.847 0.3112 1.137 1.811 0.8887
40 1.597615 62.8 2.203043 1.644814 1.174 0.1730 1.191 1.828 0.3301 1.120 1.798 0.8891
41 1.564662 61.4 2.150595 1.610227 1.166 0.1830 1.181 1.810 0.3498 1.103 1.785 0.8895
42 1.533460 60.0 2.100673 1.577454 1.158 0.1934 1.172 1.791 0.3704 1.085 1.771 0.8898
43 1.503888 58.7 2.053102 1.546370 1.149 0.2043 1.161 1.771 0.3920 1.067 1.758 0.8902
44 1.475840 57.3 2.007720 1.516862 1.141 0.2155 1.151 1.751 0.4145 1.049 1.744 0.8906
45 1.449216 56.0 1.964382 1.488829 1.132 0.2272 1.140 1.731 0.4381 1.030 1.729 0.8910
Microwave Devices, Circuits and Subsystems for Communications Engineering

46 1.423927 54.7 1.922953 1.462178 1.123 0.2394 1.130 1.710 0.4628 1.011 1.715 0.8915
47 1.399891 53.4 1.883312 1.436822 1.113 0.2521 1.118 1.689 0.4888 0.9910 1.700 0.8919
48 1.377032 52.2 1.845347 1.412684 1.103 0.2653 1.107 1.668 0.5160 0.9711 1.684 0.8923
Filters

49 1.355282 50.9 1.808954 1.389693 1.093 0.2791 1.095 1.646 0.5446 0.9508 1.669 0.8928
50 1.334577 49.7 1.774040 1.367782 1.083 0.2935 1.083 1.623 0.5747 0.9302 1.653 0.8932
51 1.314859 48.5 1.740516 1.346891 1.073 0.3084 1.070 1.600 0.6063 0.9092 1.637 0.8937
52 1.296076 47.3 1.708301 1.326965 1.062 0.3241 1.057 1.577 0.6397 0.8878 1.620 0.8942
53 1.278176 46.1 1.677322 1.307952 1.050 0.3404 1.044 1.554 0.6749 0.8661 1.603 0.8946
54 1.261116 45.0 1.647510 1.289805 1.039 0.3574 1.031 1.530 0.7122 0.8440 1.586 0.8951
55 1.244853 43.8 1.618799 1.272479 1.027 0.3752 1.017 1.506 0.7517 0.8216 1.568 0.8956
56 1.229348 42.7 1.591131 1.255935 1.015 0.3939 1.003 1.481 0.7936 0.7989 1.551 0.8961
57 1.214564 41.5 1.564449 1.240135 1.002 0.4135 0.9881 1.456 0.8382 0.7758 1.532 0.8966
58 1.200469 40.4 1.538703 1.225044 0.9894 0.4340 0.9732 1.431 0.8857 0.7523 1.514 0.8971
59 1.187032 39.3 1.513843 1.210630 0.9760 0.4556 0.9578 1.405 0.9365 0.7286 1.495 0.8976
60 1.174224 38.1 1.489825 1.196863 0.9623 0.4783 0.9420 1.379 0.9909 0.7045 1.476 0.8981
61 1.162017 37.0 1.466607 1.183715 0.9481 0.5022 0.9258 1.353 1.049 0.6801 1.456 0.8987
62 1.150388 35.9 1.444148 1.171161 0.9335 0.5274 0.9091 1.326 1.112 0.6554 1.436 0.8992
63 1.139313 34.8 1.422411 1.159176 0.9184 0.5541 0.8920 1.299 1.181 0.6304 1.416 0.8997
64 1.128771 33.7 1.401362 1.147737 0.9028 0.5824 0.8743 1.272 1.255 0.6051 1.395 0.9002
65 1.118742 32.6 1.380967 1.136826 0.8867 0.6125 0.8562 1.244 1.335 0.5795 1.374 0.9008
66 1.109208 31.5 1.361196 1.126421 0.8700 0.6445 0.8374 1.216 1.424 0.5536 1.352 0.9013
67 1.100151 30.4 1.342017 1.116505 0.8528 0.6787 0.8182 1.188 1.521 0.5274 1.330 0.9018
68 1.091555 29.3 1.323405 1.107063 0.8349 0.7153 0.7982 1.160 1.629 0.5010 1.308 0.9023
69 1.083407 28.2 1.305331 1.098078 0.8163 0.7547 0.7777 1.131 1.748 0.4744 1.285 0.9028
70 1.075391 27.1 1.287771 1.089536 0.7970 0.7972 0.7564 1.102 1.883 0.4475 1.261 0.9032
71 1.068397 26.0 1.270700 1.081425 0.7769 0.8433 0.7344 1.073 2.034 0.4204 1.237 0.9037
72 1.061511 24.9 1.254065 1.073732 0.7560 0.8936 0.7116 1.044 2.206 0.3931 1.213 0.9040
73 1.055024 23.7 1.237933 1.066446 0.7341 0.9487 0.6878 1.015 2.405 0.3657 1.188 0.9044
74 1.048925 22.6 1.222193 1.059558 0.7112 1.010 0.6631 0.9860 2.634 0.3381 1.162 0.9047
75 1.043207 21.5 1.206854 1.053059 0.6872 1.077 0.6374 0.9568 2.905 0.3105 1.135 0.9049
76 1.037860 20.3 1.191893 1.046940 0.6620 1.153 0.6104 0.9278 3.226 0.2828 1.107 0.9050

θ ωs Amin A2 A4 L1 L2 C2 L3 L4 C4 L5 C6
457

Note: (N = 6, R = 1.5, RdB = 0.18)


458

Table 6.16 Low pass prototype filter: Elliptic

θ ωs Amin A2 A4 A6 C1 C2 L2 C3 C4 L4 C5 C6 L6 C7

26 2.281172 105.4 5.038750 2.333900 2.859592 1.310 0.0290 1.358 2.100 0.1353 1.357 2.049 0.0995 1.281 1.247
27 2.202689 103.0 4.848897 2.253156 2.756829 1.308 0.0314 1.355 2.089 0.1465 1.345 2.034 0.1034 1.272 1.240
28 2.130054 100.7 4.672457 2.178409 2.661529 1.306 0.0339 1.353 2.078 0.1582 1.332 2.019 0.1117 1.263 1.233
29 2.062665 98.5 4.508037 2.109040 2.572921 1.304 0.0364 1.350 2.066 0.1704 1.319 2.003 0.1204 1.254 1.226
30 2.000000 96.3 4.354434 2.044515 2.490337 1.302 0.0391 1.347 2.054 0.1833 1.305 1.987 0.1295 1.245 1.218
31 1.941604 94.2 4.210595 1.984368 2.413194 1.299 0.0420 1.344 2.042 0.1966 1.292 1.970 0.1390 1.235 1.210
32 1.887080 92.2 4.075602 1.928190 2.340984 1.297 0.0449 1.341 2.029 0.2106 1.277 1.952 0.1490 1.225 1.202
33 1.836078 90.2 3.948647 1.875623 2.273259 1.294 0.0479 1.338 2.016 0.2252 1.262 1.934 0.1593 1.214 1.193
34 1.788292 88.3 3.829016 1.826351 2.209625 1.292 0.0511 1.335 2.002 0.2404 1.247 1.916 0.1702 1.204 1.184
35 1.743447 86.4 3.716076 1.780095 2.149731 1.289 0.0544 1.332 1.988 0.2562 1.232 1.897 0.1815 1.193 1.175
36 1.701302 84.6 3.609267 1.736606 2.093268 1.286 0.0578 1.328 1.973 0.2727 1.216 1.878 0.1932 1.181 1.165
37 1.661640 82.8 3.508087 1.695662 2.039957 1.283 0.0614 1.324 1.959 0.2900 1.199 1.858 0.2055 1.169 1.155
38 1.624269 81.0 3.412086 1.657065 1.989552 1.280 0.0650 1.321 1.943 0.3079 1.183 1.837 0.2183 1.157 1.145
39 1.589016 79.3 3.320862 1.620638 1.941830 1.277 0.0689 1.317 1.928 0.3267 1.165 1.817 0.2317 1.145 1.135
40 1.555724 77.6 3.234050 1.586220 1.896591 1.274 0.0728 1.313 1.912 0.3462 1.148 1.795 0.2456 1.132 1.124
41 1.524253 76.0 3.151325 1.553668 1.853653 1.270 0.0770 1.308 1.895 0.3666 1.130 1.773 0.2601 1.119 1.113
42 1.494477 74.3 3.072388 1.522851 1.812855 1.267 0.0812 1.304 1.879 0.3879 1.112 1.751 0.2753 1.105 1.102
43 1.466279 72.8 2.996969 1.493651 1.774048 1.263 0.0857 1.300 1.862 0.4101 1.093 1.728 0.2911 1.092 1.090
44 1.439557 71.2 2.924824 1.465961 1.737098 1.259 0.0903 1.295 1.844 0.4332 1.074 1.705 0.3076 1.077 1.078
45 1.414214 69.7 2.855727 1.439683 1.701881 1.255 0.0950 1.290 1.826 0.4575 1.055 1.682 0.3248 1.063 1.066
46 1.390164 68.2 2.789476 1.414728 1.668286 1.251 0.1000 1.285 1.808 0.4828 1.035 1.657 0.3428 1.048 1.053
47 1.367327 66.7 2.725881 1.391016 1.636211 1.247 0.1051 1.280 1.789 0.5093 1.015 1.633 0.3617 1.033 1.040
48 1.345633 65.2 2.664770 1.368471 1.605563 1.243 0.1105 1.275 1.770 0.5370 0.9944 1.608 0.3814 1.017 1.027
49 1.325013 63.7 2.605984 1.347026 1.576255 1.238 0.1160 1.269 1.751 0.5661 0.9736 1.583 0.4020 1.001 1.013
Microwave Devices, Circuits and Subsystems for Communications Engineering

50 1.305407 62.3 2.549377 1.326618 1.548208 1.234 0.1217 1.264 1.731 0.5965 0.9525 1.557 0.4235 0.9850 0.9992
51 1.286760 60.9 2.494813 1.307190 1.521349 1.229 0.1277 1.258 1.711 0.6286 0.9310 1.531 0.4462 0.9684 0.9848
52 1.269018 59.5 2.442167 1.288687 1.495612 1.224 0.1339 1.252 1.690 0.6622 0.9093 1.504 0.4699 0.9514 0.9699
Filters

53 1.252136 58.1 2.391323 1.271063 1.470934 1.219 0.1404 1.246 1.669 0.6977 0.8872 1.477 0.4948 0.9340 0.9547
54 1.236068 56.8 2.342170 1.254270 1.447259 1.213 0.1471 1.239 1.648 0.7351 0.8648 1.450 0.5211 0.9163 0.9391
55 1.220775 55.4 2.294610 1.238269 1.424533 1.208 0.1541 1.232 1.626 0.7745 0.8420 1.422 0.5487 0.8981 0.9230
56 1.206218 54.1 2.248546 1.223020 1.402707 1.202 0.1614 1.225 1.604 0.8163 0.8190 1.394 0.5778 0.8796 0.9065
57 1.192363 52.7 2.203891 1.208487 1.381735 1.196 0.1690 1.218 1.581 0.8605 0.7957 1.365 0.6085 0.8607 0.8896
58 1.179178 51.4 2.160560 1.194638 1.361575 1.190 0.1770 1.211 1.558 0.9075 0.7721 1.336 0.6411 0.8414 0.8722
59 1.166633 50.1 2.118476 1.181442 1.342188 1.183 0.1853 1.203 1.535 0.9576 0.7482 1.307 0.6755 0.8217 0.8543
60 1.154701 48.8 2.077565 1.168869 1.323537 1.177 0.1939 1.195 1.511 1.011 0.7240 1.278 0.7121 0.8016 0.8360
61 1.143354 47.5 2.037756 1.156895 1.305587 1.170 0.2030 1.186 1.487 1.068 0.6995 1.248 0.7510 0.7811 0.8171
62 1.132570 46.2 1.998983 1.145494 1.288307 1.163 0.2125 1.177 1.463 1.129 0.6748 1.218 0.7925 0.7602 0.7976
63 1.122326 44.9 1.961181 1.134644 1.271668 1.155 0.2225 1.168 1.438 1.195 0.6498 1.188 0.8369 0.7389 0.7776
64 1.112602 43.7 1.924292 1.124323 1.255641 1.147 0.2331 1.159 1.412 1.267 0.6245 1.157 0.8845 0.7171 0.7570
65 1.103378 42.4 1.888255 1.114512 1.240200 1.139 0.2441 1.149 1.386 1.344 0.5990 1.126 0.9357 0.6949 0.7357
66 1.094636 41.1 1.853014 1.105192 1.225322 1.130 0.2559 1.138 1.360 1.428 0.5732 1.095 0.9909 0.6722 0.7138
67 1.086360 39.8 1.818515 1.096346 1.210984 1.121 0.2682 1.127 1.333 1.520 0.5472 1.064 1.051 0.6490 0.6911
68 1.078535 38.5 1.784703 1.087959 1.197165 1.112 0.2814 1.116 1.306 1.622 0.5209 1.032 1.116 0.6254 0.6676
69 1.071145 37.2 1.751526 1.080016 1.183845 1.101 0.2953 1.104 1.278 1.734 0.4945 1.001 1.187 0.6013 0.6433
78 1.022341 25.1 1.472529 1.026592 1.083849 0.9782 0.4841 0.9527 1.004 3.822 0.2483 0.7148 2.368 0.3595 0.3710
79 1.018717 23.6 1.442574 1.022499 1.074724 0.9588 0.5177 0.9282 0.9699 4.337 0.2205 0.6841 2.628 0.3295 0.3316
80 1.015427 22.1 1.412537 1.018751 1.065966 0.9376 0.5562 0.9011 0.9356 4.994 0.1929 0.6540 2.946 0.2987 0.2892
81 1.012465 20.6 1.382299 1.015345 1.057569 0.9142 0.6011 0.8707 0.9006 5.858 0.1656 0.6248 3.346 0.2672 0.2431
82 1.009828 18.9 1.351718 1.012276 1.049533 0.8881 0.6545 0.8363 0.8648 7.036 0.1387 0.5968 3.863 0.2350 0.1926

θ ωs Amin A2 A4 A6 L1 L2 C2 L3 L4 C4 L5 L6 C6 L7

Note: (N = 7, R = 1, RdB = 0.18)


459
460 Microwave Devices, Circuits and Subsystems for Communications Engineering
Oscillators, Frequency Synthesisers and PLL Techniques 461

7
Oscillators, Frequency
Synthesisers and PLL Techniques
E. Artal, J. P. Pascual and J. Portilla

7.1 Introduction
In any communications system, there must be some source of the microwave signal – these
sources are generally termed oscillators. There are a number of parameters which have to be
considered when specifying a particular source, the most important being the frequency,
power, frequency stability, noise content and harmonic content.
This chapter describes the fundamentals of oscillators and frequency synthesis including
simple realisations based on diodes and transistors as well as more advanced concepts such
as voltage control of the frequency and stabilisation of the frequency using phase-locking
techniques.

7.2 Solid State Microwave Oscillators


7.2.1 Fundamentals
Oscillators are one of the main critical subsystems in microwave converters. The common
type of oscillator is based on a semiconductor active device such as a transistor or a diode
embedded in a frequency selective passive circuit. This kind of oscillator is used in practic-
ally all the low or medium power microwave systems. The semiconductor device works like
an energy exchange system: the radio frequency power delivered by the device to a load is
obtained from a DC power supply according to a specific bias point of operation.
There are several ways or approaches to analyse or to design microwave oscillators. One
possibility is to study the oscillators from the point of view of amplifiers with positive
feedback. This approach is a valuable tool in the case of low frequency oscillators but it is
not very useful in microwave applications. A more powerful and practical way for micro-
wave oscillator analysis and design is the negative resistance method. It is based on the
characterisation of the active device as a one port component with its real part of impedance
(or admittance) less than zero ohms at the operation frequency. A very simplified scheme of
an oscillator following this approach is shown in Figure 7.1.

Microwave Devices, Circuits and Subsystems for Communications Engineering Edited by I. A. Glover, S. R. Pennock
and P. R. Shepherd
© 2005 John Wiley & Sons, Ltd.
462 Microwave Devices, Circuits and Subsystems for Communications Engineering

Negative Impedance
Resistance Transforming RL
Device Network

ZD ZL

Figure 7.1 Basic scheme of a microwave solid state oscillator

I
ZD ZL

Figure 7.2 Simplified electrical circuit of a microwave oscillator

In Figure 7.1 the oscillator circuit has two separate sections, on the left the active part of
the circuit is composed of the active device (negative resistance device), on the right the
passive part of the circuit is composed of a frequency selective network including the load.
The active device can be a negative resistance diode or a transistor fed back by a suitable
passive network. The active part of the oscillator has non-linear behaviour, that is, its imped-
ance depends on the amplitude of the signal. The amplitude can be of the current or the
voltage on the active device terminal’s plane. The passive part of the device, the impedance
transforming network, is supposed to be linear, that is, its behaviour is not dependent on the
amplitude of the signal. In order to obtain the maximum radio frequency power delivered
to the load, it can be assumed that the linear network is a lossless network. In that case, the
power delivered by the active device is fully delivered to the load. In general microwave
systems the load is resistive and it is very often equal to the reference impedance system.
A standard value of this load is 50 ohm. A first analysis of the oscillator can be made from
the simplified electrical circuit shown in Figure 7.2.
The non-linear side of the oscillator is represented by the impedance ZD (device impedance)
and the linear part is represented by the impedance ZL (load impedance). This last term
must be understood as the load directly seen by the active device terminals. Assuming that
oscillations are present in the circuit, the Kirchhoff law imposes the next equation:
ZD I = −ZL I (7.1)
where I is the radio frequency current amplitude. This last equation can be now expressed
as follows:
(ZD + ZL )I = 0 (7.2)
As the current amplitude I cannot be zero, a necessary condition of oscillation is:
ZD + ZL = 0 (7.3)
Oscillators, Frequency Synthesisers and PLL Techniques 463

By solving Equation (7.3) in real and imaginary parts, the oscillation point is obtained
which is defined by the oscillation frequency f0 and oscillation amplitude I0. The practical
problem is that it is not easy to have a complete knowledge of the frequency and amplitude
behaviour of the magnitudes in Equation (7.3). In practice, there are some assumptions
that can avoid requiring the total knowledge of the active device behaviour. Moreover, the
condition of Equation (7.3) is not sufficient to guarantee the operation of the oscillator.
There is an additional condition that ensures that the oscillation point is a stable one. This
condition will be discussed later. The next example is intended to clarify some aspects of
the oscillator operation.

7.2.1.1 An IMPATT oscillator


An IMPATT diode is a special semiconductor pn junction whose impedance has a negative
resistance behaviour in a particular frequency range. This negative resistance comes from the
combination of a charge carrier avalanche (electrons and holes), due to an inverse voltage
bias of the pn junction, and a transit time effect of these carriers through the semiconductor.
A simplified equivalent circuit of the IMPATT diode, only valid for RF purposes, is shown
in Figure 7.3.
The negative resistance is non-linearly dependent on the RF current amplitude. The cap-
acitance can be approximated by the pn junction capacitance value under the reverse bias
condition. A typical negative resistance dependence is shown in Figure 7.4.

R D(I)

I
C

Figure 7.3 Equivalent circuit of an IMPATT diode

RD(OHM)

IP(A)
0.8

Figure 7.4 IMPATT negative resistance versus RF current


464 Microwave Devices, Circuits and Subsystems for Communications Engineering

The negative resistance non-linear function can be approximated by a Van der Pol character-
istic, which is based on a RF voltage cubic dependence versus the RF current:
VNL = −aI + bI 3 (7.4)
In this equation VNL is the non-linear voltage on the non-linear resistance RD(I) terminals,
and I is the peak value of the RF current across it. From Equation (7.4) the non-linear
resistance can be obtained as:
RD(I) = −a + bI2 (7.5)
A good fitting of the curve shown in Figure 7.4 is obtained selecting appropriate values for
constants a and b:
a=4 b = 3.1
A very simplified circuit to analyse an oscillator using this IMPATT diode is shown in
Figure 7.5. In this example the oscillator circuit is a series resonant type. In practice, it is not
possible to mount such a simple circuit because it is very difficult to manufacture an ideal
coil (L) and an ideal resistance load (RL ) at microwave frequencies. However, this simple
circuit is very useful to model actual oscillator circuits in practice because, very often, in
narrow frequency bandwidths, a microwave oscillator can be represented by a series resonant
circuit or by a parallel resonant circuit.
It must be noted that in Figure 7.5 the non-linear impedance is restricted to the IMPATT
diode non-linear resistance RD, while the capacitance C, which accounts for the space charge
capacitance of the reverse biased pn junction, can be considered linear. In order to identify
both the linear and non-linear parts of the oscillator circuit, the capacitance C can be
included on the right side of the circuit as a load subcircuit element. This is identified in the
figure as ZD (device impedance) and ZL (load impedance).
Usually the oscillator is designed to obtain the maximum oscillation power from the active
device. The value to be maximised is the power P delivered to the load:
1 1
P= RLI 20 = − (−a + bI 20)I 20 (7.6)
2 2

ZD ZL

C L

I
R D(I) RL
1
4
4
2
4
4
3
3
4
4
2
4
4
1
3
4
4
2
4
4
1

Diode Impedance Load


Transforming
Network

Figure 7.5 Series resonance type IMPATT oscillator circuit


Oscillators, Frequency Synthesisers and PLL Techniques 465

In order to find the optimum point, the derivative of power P versus the RF current peak
amplitude I0 must be zero:

∂P
= aI0 − 2bI 03 = 0 (7.7)
∂I0

From this condition the optimum RF current amplitude is obtained:

a
I0 = ≅ 0.8 (A) (7.8)
2b

As a consequence of the last equation, the optimum non-linear resistance value is found:

a
RD(I0) = − = −2 (ohm) (7.9)
2

This result is important because can be interpreted as follows: a non-linear negative resistance
device, with Van der Pol voltage to current characteristic, will deliver the maximum power
to a load if its resistance has a value half that of its low signal resistance value. The low
signal resistance value is equal to –a in this example. The low signal value occurs at very
low levels of current (or voltage). From Equation (7.5) the low signal resistance value RD(0)
is obtained assuming the value of I is near zero:

RD(0) = limI→0[RD(I)] = −a (7.10)

Many practical devices have Van der Pol characteristics, such as most negative resistance
diodes, so the Equation (7.9) approach is very useful because the device impedance measure-
ment under low signal level conditions is usually an easy task. On the other hand, the device
impedance measurement under large signal conditions can be a very difficult task due to the
special test equipment required.
From Equation (7.3) the load resistance RL for maximum oscillation power must be equal
to the device optimum non-linear resistance but with opposite sign:

a
RL = = 2 (Ohm) (7.11)
2

The power delivered to this load is then:

a2
P= ≅ 645 mW (7.12)
8b

The oscillation frequency is determined by the series resonance, see Figure 7.5, so it can be
obtained from the next identity:

1 1
f0 = (7.13)
2π LC

The signal at the output load is a sinusoidal one with frequency f0 (Hz) and with an average
power P.
466 Microwave Devices, Circuits and Subsystems for Communications Engineering

7.2.2 Stability of Oscillations


The condition of oscillation given by Equation (7.3) is a necessary condition to have oscilla-
tion in a circuit, but this condition is not enough to have a steady sinusoidal signal in the
load. There are several ways to analyse the stability of the oscillation point. At microwave
frequencies there is a useful graphical method to do the stability test. The method is based
on the definition of two impedance lines called the device line and the load line. The device
line is the graphical representation of the device impedance with a change of sign. The
device impedance is in general dependent on two variables: frequency and signal amplitude.
In a strict sense the device line is not a line but a surface. In practice, the device impedance
is a slowly varying function against frequency if it is compared with the stronger variation
of load impedance against frequency. Following the data from the last example (p. 000),
the device line can be defined as the graphical representation of −ZD(I) and the load line as
the graphical representation of ZL( f ).

Device line −ZD(I, f ) ≅ −ZD(I)


Load line ZL( f )

From the data of the IMPATT oscillator example, both lines can be drawn on the impedance
plane Z = R + jX. These lines are shown on Figure 7.6. The point where the lines cross defines
both parameters of the oscillator: current amplitude I and frequency of oscillation f0.
In the general case, neither lines have an easily defined analytical expression. In this
example, due to its simplicity, it is possible to write:

Device line −ZD(I) = 4 − 3.1I 2


Load line ZL( f ) = RL(1 + j2Qδ )

With

a f − f0
RL = =2 Q = ω 0 L / RL δ =
2 f0

which are respectively the load resistance, the quality factor of the resonant circuit and the
relative frequency shift.

X
Z L(f)

0 2 4 R

–Z D(I)

–Z D(0)

Figure 7.6 Device and load lines for the oscillator circuit of the IMPATT oscillator
Oscillators, Frequency Synthesisers and PLL Techniques 467

Classical stability analysis of oscillations can be done assuming small amplitude and
frequency deviations from the oscillation point given by I0 and f0. If the system reacts so
that the deviations decrease with time, in such a case the oscillation point is a stable one.
However, if the deviations increase with time, for this point a stable oscillation is not
possible. A method of analysing stability conditions is based on the properties of impedance
functions. These are complex variable analytical functions, so some interesting properties
can be extracted. A mathematical study of such functions is beyond the scope of this book,
and only the final graphical condition for stable operation is presented. A complete analysis
is included in [1]. Using the graphical representation of device and load lines, the stability
condition of oscillations can be expressed as follows: ‘The measured angle from the device
line to the load line, in the growing sense of arrows, must be less than 180° to have stable
oscillator behaviour’. An example is shown in Figure 7.6.

7.3 Negative Resistance Diode Oscillators


Two terminal semiconductor devices were the first solid state devices used by the micro-
wave industry. Such devices were developed in the 1960s. Nowadays some of these are
available in most microwave product catalogues, using standard commercial names such as
Gunn diodes or IMPATT diodes, which are the more extensive types. A common charac-
teristic of these devices is that their impedance presents negative resistance behaviour in a
limited frequency range. The physical origin of their negative resistance (or admittance) is
very different in Gunn diodes compared with IMPATT diodes. A description is included
in Chapter 2.
At present, negative resistance diode oscillators are the more powerful solid state sources
at very high frequencies, and the only solid state source possible in the higher frequency
range of millimetre waves. Most microwave diode oscillators are built in coaxial or waveguide
structures, more precisely coaxial resonators or waveguide cavities, due to their high quality
factors. A high quality factor of the load circuit implies good frequency stability and low
noise. It is possible also to build diode oscillators using microstrip lines or other similar
printed transmission lines, but in this case some additional circuitry must be added to achieve
good frequency stability.
From the designer’s point of view, the knowledge of the device impedance as a function
of frequency and signal level is enough. Unfortunately often one only has some limited data
for the device. Some manufacturers give on their data sheets the optimum impedance values
for maximum output power. Often a difficulty in the design of such oscillators arises from
the relative low resistance levels of diodes. Those values are in general lower than 10 Ohm
while the standard system impedance is 50 Ohm. Moreover, the waveguide mountings
have inherently high impedances, so special mounting procedures are used to obtain high
transformation ratios avoiding resistive losses as much as possible. The reactive part of
device impedance tends to be very high, especially in the IMPATT diode case, making the
transformation impedance network a crucial issue in the oscillator design. Diodes in chip
version have a capacitive behaviour but bonding and other package effects can transform
this to an inductive behaviour.
The efficiency of negative resistance diodes is very low, so the majority of the DC power
delivered to the device is dissipated inside. The devices are normally available in a packaged
version. Packages are designed to act as heatsinks to reduce the internal temperature of the
468 Microwave Devices, Circuits and Subsystems for Communications Engineering

ø 1.50 mm

Wires

Ceramic

0.25 mm
Ga As Chip

Heatsink

Figure 7.7 Typical packaged Gunn diode

Lp

Cp Gd Bd

Figure 7.8 Simplified equivalent circuit of a packaged Gunn diode

semiconductor to a reliable operating value. Package parasitic effects produce significant


impedance changes, so the package type must be carefully selected to fulfil both heatsinking
and impedance transformation requirements. A typical packaged Gunn diode is shown in
Figure 7.7. The semiconductor chip is inside, the top contact connections are made by wire
bonding. Diode electrodes are electrically isolated by a cylinder of ceramic material.
A simplified equivalent circuit of the active device (Figure 7.8), is composed of a negative
conductance Gd and a shunted capacitive susceptance Bd. Packaging effects are included as
a series inductance Lp and a shunt capacitance Cp.
Typical values for a Gunn diode like the example shown in Figure 7.7 are:

Lp = 0.1 nH Cp = 0.2 pF

Non-linear behaviour of the conductance Gd and susceptance Bd are useful to analyse or


design the oscillator performance. Unfortunately these data are difficult to obtain from tests
and most often the designer must undertake the design without it.
Oscillators, Frequency Synthesisers and PLL Techniques 469

7.3.1 Design Technique Examples


There are three basic ways to build Gunn oscillators: (1) in a coaxial cavity; (2) in a wave-
guide; and (3) in a microstrip line. Figure 7.9 shows an example of each. In the coaxial cavity
case, the oscillation frequency is determined by the cavity length l. It is approximately a half
wavelength (l ≅ λ /2). The Gunn diode is located approximately at a λ /8 distance from the
end of the cavity. A frequency adjustment can be added by inserting a dielectric or metallic
screw, located at a point with maximum electrical field to have maximum mechanical adjust-
ment range. Power output can be adjusted by variation of the inductive coupling loop.
The main disadvantage of coaxial Gunn oscillators is their low quality factor, a typical
value is 50, which produces poor frequency stability for most applications. Special care must
be taken to avoid capacitive discontinuities in the junction diode to coaxial inner conductor.
A suitable inner conductor diameter must be chosen. Other criteria to select the outer con-
ductor diameter are: at oscillation frequency higher coaxial propagation modes must be
under cut-off, the diameter ratio must be selected to have a maximum value of the cavity
unloaded quality factor. Biasing from a DC power supply is obtained by electrical separation
of the diode electrodes using a radial line of a quarter wavelength, so in RF it is equivalent
to a short circuit at both coaxial ends. The dielectric separating both radial line conductors
can be a very thin Mylar® sheet.
A possible waveguide assembly is shown in Figure 7.9(b). It represents a longitudinal
view of a rectangular waveguide structure and it is an iris coupled oscillator. The oscillation
frequency is approximately determined by a half wavelength distance from the iris to the
diode position. Tuning is possible by a screw insertion, tuning ranges up to 30% are feasible.
The Gunn diode is coupled to the cavity by a cylindrical post. It also allows introduction
of the DC polarisation through a coaxial low pass filter. This filter has quarter wavelength
sections alternating low and high characteristic impedances. Waveguide assemblies like this
one can give high quality factors, around 1000, and hence high stability Gunn oscillators.

7.4 Transistor Oscillators


The basic principle of any electronic sinusoidal oscillator is in the use of the resonance
phenomena. It is the case, for instance, that in the application of some initial energy to an
ideal LC resonant circuit it produces an oscillating sinusoidal signal as the inductor and the
capacitor periodically interchange this energy. Nevertheless, in the real case, due to the losses
in the inductor and capacitor, the oscillating signal will decay exponentially with time. To
maintain the oscillation, an active device must be used in order to compensate for this energy
loss. For this purpose, transistor oscillators are widely used in radio frequency and micro-
wave systems, offering good performance and high integration with other subsystems built
using transistors. The choice of a particular transistor, as well as the resonant structure and
the oscillator topology, will depend on the oscillating frequency to be obtained and on the
particular application of the oscillator. In general, bipolar transistors are commonly used at
radio frequencies and in the lower microwave bands (up to a few GHz). Above these fre-
quencies, GaAs field-effect transistors are employed because of their ability to work at higher
frequencies. Moreover, due to the development of device technology, HBTs (Heterojunction
Bipolar Transistors) and HEMTs (High-Electron Mobility Transistors) are also available and
have demonstrated good performance in microwave and millimetrewave oscillators.
470 Microwave Devices, Circuits and Subsystems for Communications Engineering

λ/8
Gunn Diode
Tuning Screw

+ Output
Coupling

λ/4
Radial
Line
(a)
Tuning +
Screw
Low Pass Filter

Post

Iris

Gunn Diode

Assembly
λg/2 Mount
(b)

Vertical
+V Ground Block
Low
Pass
Filter Packaged
Gunn Diode

DC Block Transformer
(c)

Figure 7.9 Gunn diode oscillators: (a) coaxial mount; (b) waveguide cavity; (c) microstrip
Oscillators, Frequency Synthesisers and PLL Techniques 471

The design techniques of transistor oscillators are, in general, independent of the device
technology. However, the achievable performances, in terms of output power or phase noise,
for instance, can be strongly dependent on it. Nevertheless, these aspects are beyond the
scope of this text, we restrict ourselves here to exploring the design fundamentals of transis-
tor oscillators, showing the more common circuit topologies and illustrating them by means
of some practical examples.

7.4.1 Design Fundamentals of Transistor Oscillators


A microwave transistor, as shown in Chapter 2, is a semiconductor three-terminal device,
which acts, mainly, like a dependent current source. The performance of a given transistor is
determined, avoiding second-order and high-frequency parasitic effects, from the character-
istic curves of this current generator. In the case of bipolar transistors, the current source is
controlled by the base current and the collector-emitter voltage, whereas in unipolar transistors,
such as MESFET or HEMT, the gate-source and drain-source voltages are the controllers of
the current source. Under a proper selection of the transistor bias conditions, that is, the DC
operating point, the transistor is able to amplify a signal applied to its input terminal.
Generally speaking, a transistor oscillator can be seen as a transistor amplifier having
positive feedback, allowing the growth of any starting oscillating signal in the circuit, coupled
to a resonant circuit, which serves to select the frequency of the oscillation. In order to start
up the oscillation, a signal containing a spectral component at the desired frequency must be
available in the oscillator circuit. However, this initial signal is already present in all the
electronic circuits, in the form of noise. White noise, for instance, is produced in any circuit
component due to the fact of it being at a given temperature. This thermal noise is called
white because the associated spectrum is frequency independent.
In the oscillator, the white noise within a frequency band is amplified and a portion of the
amplified signal, at the frequency determined by the bandpass characteristic of the resonator
circuit, is fed back to the input. When the loop gain is unity and the phase shift between the
output and input signals is 360°, the oscillation will be possible, known as the Barkhausen
condition for oscillation.
Under such conditions, the oscillating signal would continuously grow until it reached
a level determined by any mechanism of signal limitation present in the circuit. These
mechanisms can be related to the physical amplification limits of the transistor itself, pro-
duced when the oscillating signal reaches the maximum voltages and currents achievable
in the characteristic curves. It can be also introduced externally by adding signal limitation
circuitry into the oscillator loop. In this text, we will only consider the former kind of signal
limitation, which is in fact common in general-purpose transistor oscillators. When the final
stable oscillation condition is reached, the losses associated with the passive circuitry and
the negative resistance provided by the active device, dependent on the signal magnitude,
have equal contributions. Following this baseline, a schematic of the representation of a
generic transistor oscillator is shown in Figure 7.10.
The design procedure of a transistor oscillator, as can be seen from the discussion above,
is quite close to the amplifier design. Remember that, in a practical small-signal amplifier
design, a reference terminal is usually grounded and specific impedances, calculated to
satisfy the matching conditions at the frequency of the amplified signal, are connected at the
input and output ports. Furthermore, amplifiers can also use negative feedback in order, for
472 Microwave Devices, Circuits and Subsystems for Communications Engineering

A = µ /(1 – µβ )
µ

Figure 7.10 Feedback oscillator scheme

instance, to reach better stability or increase the amplification bandwidth. When compared to
the design of a microwave small-signal amplifier, the main difference in the design of the
transistor oscillator is in the need to create a positive feedback path and sufficient negative
resistance to make possible the start and growth of the oscillating signal.

7.4.1.1 Achievement of the negative resistance


One of the key points in the design procedure of a transistor oscillator is how to obtain
sufficient negative resistance, in the desired oscillating frequency range, to compensate for
the losses of the passive circuitry of the oscillator, thus permitting the start of the oscillation.
First of all, the achievement of the negative resistance requires biasing the transistor in
the active or amplification region. Second, sufficient impedance should be connected to the
transistor terminals in order to produce either series or parallel (or both) feedback. This
involves the selection of a circuit topology and the calculation of the values of the different
components. Finally, the resonator part must be connected or coupled somewhere into the
circuit, in order to tune the desired oscillating frequency. In the following paragraphs, we
will deal with different oscillator circuit topologies and choices for the implementation of
the resonator part.
The generic topology of an oscillator using a transistor as the active device can be
represented as shown in Figure 7.11. In this figure, the three-terminal device represents the

Figure 7.11 Generic topology of a transistor oscillator


Oscillators, Frequency Synthesisers and PLL Techniques 473

transistor and Z1, Z2, Z3 and Z4 are, in general, complex impedances. From this circuit
scheme, different configurations of transistor oscillators can be produced. Series feedback
(Z3 = ∞), parallel feedback (Z2 = 0) or a combination of both (Z2 ≠ 0, Z3 ≠ 0) are possible
in such a configuration.

7.4.1.2 Resonator circuits for transistor oscillators


Before exploring the usual resonator circuits employed in radio frequency and microwave
oscillators, let us give a short review of some basic ideas about the frequency selective
properties of these kind of circuits. Remember that the losses in a resonant circuit produce a
transfer function in the frequency domain having a band pass characteristic instead of a flat
frequency response. The band pass frequency is commonly defined as being the difference
between the upper and lower frequencies at which the magnitude of the transfer function
is 3 dB below the response in the band. The bandwidth is inversely related to the quality
factor Q, which acts as a measure of the frequency selectivity of the resonant circuit. It can
be written:

ω0
Q= (7.14)
ω2 − ω1
where ω 0 is the centre frequency and ω 2 and ω 1 are the upper and lower frequency bandpass
limits. It is obvious that the higher Q is, the narrower the bandwidth and the higher the
frequency selectivity of the circuit.
The Q factor can be determined as a function of the values of the inductor, capacitor and
the resistive losses. The expressions, for series and parallel resonant circuits (as illustrated in
Figure 7.12) are:

ω0 L 1
QS = = (7.15)
R RCω 0

R
QP = = RCω 0 (7.16)
ω0 L
Furthermore, it is important to remember that, if an external load is connected to the resonant
circuit, the overall circuit Q (known as loaded Q or QL) must take into account the additional
losses due to the external circuit. This is because, for a given resonator and load impedance,
the QL of the circuit can be very different from the Q of the resonator circuit itself. The

R L C C
a/series
b/parallel
Figure 7.12 Series and parallel resonant circuits
474 Microwave Devices, Circuits and Subsystems for Communications Engineering

expression relating the QL with the circuit Q and the Q associated to the external losses only,
QEXT, is:

1 1 1
= + (7.17)
Q L Q QEXT

Different options can be adopted to implement the passive part of the oscillator, i.e. the
feedback and the biasing and tuning circuits. Capacitors and inductors are available to
construct resonant circuits at high frequencies, having high enough quality-factor Q and
stable behaviour with temperature drift. The resonant circuits implemented with inductors
and capacitors are intensively applied in the design and construction of radio frequency
oscillators for different electronic systems. This solution gives the so-called L-C oscilla-
tors in which the oscillating frequency is related to the resonant frequency of the L-C
resonator.
Other alternatives have been adopted in the design of the resonator circuits for high fre-
quency oscillators. The use of transmission lines instead of lumped L-C circuits is common
in the design of microwave oscillators. This it is because of the increasing parasitic effects in
the inductors and capacitors with the frequency, lowering the Q and often exhibiting undesired
resonant frequencies under or near the desired oscillation frequency. The implementation
of the passive circuits of the oscillator using transmission lines permits the easy and low
cost integration of the oscillator. Usual resonators implemented with transmission lines can
be built using coaxial line or microstrip line. Such examples are shown in Figure 7.13. For
instance, cavity resonators made with open circuit or shorted coaxial lines are commercially
available up to the lower microwave bands.
The microstrip line resonators can be made using shorted or open circuit transmission
lines or other planar structures such as the ring resonator.
Very high-Q resonators at microwave frequencies can be obtained from circular or rectan-
gular waveguides. Low-cost high-Q resonators can be obtained using dielectric resonators
coupled to microstrip transmission lines, giving very good performance at microwave fre-
quencies. Examples are shown in Figure 7.14. All these resonator types are usually employed
for low noise fixed frequency oscillators but also can be used to implement mechanically
or electronically tuned oscillators, with some extra circuitry added. Specific resonators for
electronically tuned oscillators will be presented later in this chapter.

λ/4 λ/2
(a) (b)

Figure 7.13 Resonators using transmission lines: (a) λ/4 resonators; (b) λ/2 resonator
Oscillators, Frequency Synthesisers and PLL Techniques 475

Rr

Cr
Lr

Zo Rj Lj
Cj Zo

Figure 7.14 Schematic of resonator coupled to a microstrip line

7.4.2 Common Topologies of Transistor Oscillators


Some particular topologies of transistor L-C oscillators have been extensively employed in
electronic systems for different applications. These are harmonic or sinusoidal oscillators,
in which the oscillation frequency is fixed by the values of the inductors and capacitors in
the circuit. There are other ways to produce an oscillation, as in the relaxation oscillators,
associated with the relaxation time of the charging and decharging processes in a capacitor
through a resistor. Nevertheless, we will focus our attention on harmonic oscillators because
they provide the possibility of achieving higher frequencies (because no relaxation time is
involved in the generation of the oscillation) with higher Q (because resistors, introducing
extra losses, are not needed to produce the oscillation).
Among the several possibilities in implementing L-C oscillators, we will explore the
Colpitts, Clapp and Hartley topologies. All these solid-sate oscillators are, in fact, an evolu-
tion from oscillators originally built using feedback vacuum-tube amplifiers. The differences
between them arise from the system employed to feed back a part of the energy from the
resonator and in the implementation of the resonator itself (see figures later in the chapter).
For instance, both the Hartley and Colpitts structures use a parallel resonator but, whereas
in the former an inductive voltage divider is used to feed back the signal, the Colpitts
employs a capacitive voltage divider. In its turn, the Clapp oscillator also uses the capacitive
feedback but introduces a series resonance in the inductive branch. Furthermore, different
versions of the topologies above can be obtained depending on the AC grounded terminal
considered. This means that we can talk about the grounded-emitter, grounded-collector
or grounded-base versions (grounded-source, grounded-drain and grounded-gate when using
MESFETs or HEMTs), even if, in a strict sense, the common-emitter, common-collector or
common-base terms are used, as in the case of amplifiers, when an input signal is applied
to the circuit. For such reasons, even if sometimes it is difficult to identify them, most of the
radio frequency oscillators are, in fact, for instance, Colpitts or Hartley oscillators, or slight
variations of these.
In the following sections, we will study in more detail these LC oscillators, but also some
examples of microwave transistor oscillators, implemented using transmission lines instead
of lumped elements, will be presented.
476 Microwave Devices, Circuits and Subsystems for Communications Engineering

C1
L

C2

Figure 7.15 Schematic of the Colpitts oscillator

7.4.2.1 The Colpitts oscillator


Figure 7.15 shows the schematic of the Colpitts oscillator. The transistor is used in grounded-
base configuration (Z1 = 0 in Figure 7.11) and acts simultaneously as the amplifier and
signal limiter blocks in the oscillator loop. The frequency-tuning block is a parallel L-C
resonator connected to the drain terminal (Z3 in Figure 7.11). The resonator capacitor is, in
fact, substituted by a capacitive divider, which is used to feed back a part of the signal to the
input of the gain block. It is possible to get a first estimate value of the oscillation frequency
from the values of the resonator elements, as given in the following expression:

1
ω0 = (7.18)
CC
L 1 2
C1 + C2

The loaded Q in this oscillator configuration increases by increasing C2 and decreasing C1


in the capacitive divider, or by increasing both capacitor values and decreasing the inductor
L. The output signal is usually taken from the resonator output, the source terminal of the
transistor when using a MESFET device, because of the better spectral purity. A coupling
technique or a buffer amplifier can be employed in order to extract the signal through the
output load.
More accurate analytic calculations of the oscillation frequency for a practical example
of a Colpitts oscillator implemented using a MESFET transistor and including two buffer
amplifiers which are used to extract the oscillating signal for a 50 Ω load and a prescaler,
for instance, are presented. The analytic expressions are obtained by using the equivalent
circuit of the transistor and by applying the Barkhausen oscillation condition (voltage
gain around the loop equal to unity, with null phase). In Figure 7.16, the configuration of
the Colpitts oscillator used in the calculations is shown. The oscillation frequency can
be written:

Cds + C1 + C2′
ω0 = (7.19)
L[(Cbuffer + C2′ ) + (Cds + C1 ) + (CbufferC2′ )]

The constraint in the gain is:

1 + gm Rds
=1 (7.20)
(ω L)(gm RdsCbuffer + Cbuffer + C2′ )
2
Oscillators, Frequency Synthesisers and PLL Techniques 477

Cds
s
C1
LC
2

s Cds
Cgs
gm
Cbuffer Cbuffer
Rds

Figure 7.16 Schematic of the Colpitts oscillator used in the calculations

C1
L

C2 C3

Figure 7.17 Schematic of the Clapp oscillator

In these expressions C′2 = C2 + Cds+ Cgs + Cbuffer and Cds, gm and Rds are the elements of the
transistor small-signal model, Cbuffer corresponds to the buffer input capacitor and C1 and C2
are the capacitors of the capacitive divider.

7.4.2.2 The Clapp oscillator


The Clapp structure is quite similar to that of the Colpitts oscillator. The difference, in
this case, lies in the series connection of a capacitor to the resonator inductor. The Clapp
schematic is shown in Figure 7.17. The advantage over the Colpitts oscillator is of the
higher loaded Q obtained for a given inductor value. The increase of the inductor value
and the decrease of the in-series capacitor result in the increase of the loaded Q. As in
the Colpitts oscillator, the loaded Q can also be increased by reducing the C1 value and
increasing C2.

7.4.2.3 The Hartley oscillator


Another example of an oscillator based on the use of an amplifier with feedback, is the
Hartley topology. As well as in the Colpitts structure, a parallel resonator is employed as the
frequency-tuning circuit, but the sample of the signal to be fed back is obtained from an
inductive voltage divider, instead of using the capacitive divider as in the Colpitts oscillator.
The Hartley scheme is shown in Figure 7.18.
478 Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 7.18 Schematic of the Hartley oscillator

7.4.2.4 Other practical topologies of transistor oscillators


Figure 7.11 summarised the generic topology of a transistor oscillator. A method for design-
ing an oscillator from this scheme is based on the selection of appropriate impedance values
for Z1, Z2, Z3 and Z4. By making use of a simplified small-signal equivalent circuit of the
transistor, for the desired bias point in the amplification region of the transistor characteristics,
simple expressions for the preliminary design of a transistor oscillator can be obtained for
a particular configuration. For instance, a practical design is obtained by assuming only a
capacitive series feedback (a capacitor in Z2, Z3 = ∞) and an inductor in Z1. In this case, the
negative resistance seen towards the transistor from the Z4 port can be written:

− gm
Ro = (7.21)
Cdsω (Cgs + C2 − L1C2Cgsω 2 )
2

Making use of the oscillation condition at the Z4 port (Zosc = Z4) and assuming that Z4 is
a resistive impedance, the oscillation frequency can be calculated:

Cds + Cgs + C2
ωo = (7.22)
L1Cgs (C2 + Cds )

Other alternatives can be adopted by using an LC series resonant circuit connected to the
gate, parallel feedback, etc. Some of these topologies can be, in fact, like the designs already
shown above.

7.4.2.5 Microwave oscillators using distributed elements


In the previous subsections we have seen some examples of oscillators using transistors
with positive feedback, in which the feedback and resonator circuitry was implemented
with lumped elements such as inductors and capacitors. It is also possible to build transistor
oscillators at higher frequencies having similar structures but using transmission lines instead
of lumped elements.
Considering the transistor as a two-port device, two basic topologies, with external networks
implemented with transmission lines in Π and T topologies, can be used to build shunt or
series oscillators respectively (see Figure 7.19).
A practical solution in achieving high-Q microwave transistor oscillators is the use of
dielectric resonators coupled to a transmission line into the oscillator circuit. These kinds
of oscillators provide a very stable oscillating signal, with high spectral purity. A topology
widely employed is by coupling the resonator to the transistor base or gate, as shown in
Oscillators, Frequency Synthesisers and PLL Techniques 479

Transistor Transistor

Figure 7.19 π and T topologies using transmission lines

s s

Figure 7.20 Dielectric resonator oscillators

Figure 7.20. Sometimes a microwave oscillator can use a dielectric resonator in order to
decrease the phase noise, but also a voltage-controlled capacitor to achieve frequency-tuning
in a narrow frequency band around the oscillation frequency set by the resonator.

7.4.3 Advanced CAD Techniques of Transistor Oscillators


The analytic approaches to the design of oscillators can be very useful to analyse a given
oscillator topology and to get starting values for the main elements of the oscillator, offering
quite good results at low frequencies. Nevertheless, because of the assumptions needed
in order to obtain simple analytic expressions, the accuracy of the predicted performances
will be limited. Some issues related to these limitations and design techniques that help
to avoid these problems will now be introduced. It is important to note that the analytic
designs in the previous paragraphs have been performed in the small-signal region and
thus are only useful when the oscillating signal is starting, after the transistor biasing. The
achievement of more accurate results of the negative resistance and the oscillating frequency
in the small-signal region requires more complex calculations or simulations, taking into
account the passive circuit parasitics, as well as a more realistic small-signal model of the
transistor.
An approximate method for obtaining the maximum output power of a transistor oscilla-
tor was analytically derived by Johnson [2]. It makes use of the small-signal gain and the
saturated power achievable using the same transistor in a large-signal amplifier design:

 1 ln(G0 ) 
PMAX = PSAT 1 − −  (7.23)
 G0 G0 

where PSAT is the saturated output power in the amplifier and G0 is the small-signal common-
source transducer gain of the amplifier.
480 Microwave Devices, Circuits and Subsystems for Communications Engineering

Simulation techniques have an important role as a solution to minimise the production


cost and time to market, providing useful information to aid the design of the oscillator.
Commercial simulators use analysis techniques offering small-signal oscillation test or analysis
in the non-linear region such as transient techniques or harmonic-balance methods. The
small-signal test is performed by breaking the oscillator loop and by inserting an s-parameter
port, giving the reflection coefficient of the oscillator at that port. Oscillation is possible if
the magnitude of the given reflection coefficient is greater than one with null phase. These
kinds of analysis serve to guarantee the starting condition of the oscillation. In order to be
able to make a prediction of the final frequency and power of the oscillating signal, when
the steady state of the oscillation is reached, the non-linear behaviour of the transistor must
be known and a non-linear analysis technique must be used.
Harmonic balance is widely employed in the design of RF and microwave circuits because
of the low computing time requirements compared to time-domain techniques, which need
the calculation of the transient response in small time steps for the achievement of the
steady-state response of the circuit. The harmonic-balance method divides the circuit in
two parts: the passive part, described in the frequency domain, and the active part, described
using expressions in the time domain of the different sources controlled by the correspond-
ing command voltages or currents. The method converges producing the balance, at every
working frequency and its harmonics, at the ports connecting both sub-circuits. In order to
obtain the oscillation frequency, an auxiliary source, with varying frequency and amplitude,
is used in several iterations until the oscillation is self-supported. The solution provides
information about the power levels at the fundamental frequency and its harmonics, giving
the steady state but not any guarantee of the start of the oscillation. In contrast, the time-
domain methods offer the time evolution of the oscillating signal, from the starting point to
the steady state.

Self-assessment Problems
7.1 Evaluate the oscillation frequency in a Clapp oscillator, as a function of the values
of the passive components in the circuit.

7.2 Evaluate the oscillation frequency in a Hartley oscillator, as a function of the


values of the passive components in the circuit.

7.3 A bipolar transistor is employed in the design of an oscillator, working at 1 GHz,


whose schematic is in the Figure Q3. The scattering parameters for such transistor,
corresponding to the working bias point and frequency are:

S11: (0.9, −100°), S12: (0.5, 31°), S21: (1.1, −50°), S22: (0.6, 150°)

Calculate the device impedance ZD = RD + jXD seen at the transistor base and the
values of L and ZO, corresponding to the load impedance ZL, in order to obtain an
oscillating signal at 1 GHz, considering that Re(ZL) = RD /3 is the optimum value
for obtaining the desired oscillation power using the given transistor.
Oscillators, Frequency Synthesisers and PLL Techniques 481

l = λ/4 L?

C = 16 pF
50 Ω Z O?

ZL ZD

Figure Q3 Schematic of the oscillator, showing the device and load impedances ZD and ZL

7.5 Voltage-Controlled Oscillators


Oscillators having a frequency-tuning facility are in demand in high-frequency systems for
consumer and professional applications. Because the oscillation frequency is basically deter-
mined by the resonant circuit, the frequency tuning can be achieved by varying its resonant
frequency. The frequency tuning mechanism can be mechanical or electrical, depending on
the type of resonator used. In the former case, a lumped-element resonator or a cavity resonator
can be used in order to obtain a tuning bandwidth by, for instance, varying the value of a
trimmer capacitor or by changing the length of the cavity, respectively. YIG resonators or
varactor diodes could be the choice if an electrical frequency tuning is required.
The YIG (Yttrium Iron Garnet or ferrite sphere) or the varactor resonators give the
possibility of wideband electrically tuneable oscillators. The YIG is a high-Q resonator in
which the ferromagnetic resonance depends on the material, size and applied field. The
frequency can be tuned over a wideband by varying the biasing of the magnetic field across
the ferrite. The YIG oscillators are used in applications requiring very high quality tuneable
oscillators, such as in microwave sources up to 60 GHz for instance.
When the circuit size or the cost is an important issue, if lower Q is acceptable, the choice
is to use voltage-tuned varactors as frequency tuning elements in the microwave oscillators.
It is for this reason that varactor-tuned oscillators are present in almost all the commercial
applications and why, in this text, we will restrict ourselves to the study of varactor-tuned
oscillators.
Varactors are special diodes showing a wide range of voltage-controlled variable capacit-
ance. This is the key aspect for achieving broadband frequency-tuning capability. Another
important issue is the minimisation of the varactor series resistance, in order to increase the
Q factor. Silicon hyperabrupt varactors show capacitance ratios (Cmax/Cmin) greater than 12.
Gallium-Arsenide varactors, based on Schottky diodes, are also employed, showing higher
Q values because of the lower series resistance associated with the metal-semiconductor
junction compared to the pn junction varactors realised in silicon.

7.5.1 Design Fundamentals of Varactor-Tuned Oscillators


As stated before, varactor-tuned oscillators use the voltage-capacity-varying characteristic
of the varactor diodes, in order to obtain electronically frequency-tuning capability. In
482 Microwave Devices, Circuits and Subsystems for Communications Engineering

principle, by simple substitution of the capacitor in the resonator of any given L-C oscillator
topology with a varactor, frequency tuning is possible with the varactor bias voltage. Never-
theless, at this point, it is important to remember that the parasitics in the varactor diode
and other circuit elements will introduce correcting terms in the resonant frequency and also
will decrease the Q factor of the resonant circuit. The deviation of the capacitor dependence
with voltage from the expression of the ideal diode, produces a non-linear variation of the
resonant frequency with the tuning voltage.
In order to illustrate how a varactor can serve as tuning element, let us consider a
series resonant circuit in which the capacitor has been substituted by a varactor. The
relationship between the varactor capacitance and the voltage Va applied can be written,
for Va < 0 V:

Cj 0
C= γ
(7.24)
 Va 
1 − 
 φB 

where Cj0 depends on the varactor’s physical and geometrical parameters, φB is the barrier
potential of the junction and the γ value depends on the doping profile. Using this, the
resonant frequency versus the applied voltage Va, is:
γ /2
 Va 
1 − 
 φ
fR = (7.25)
2π LC j 0

When γ ≅ 2, as in the case of hyperabrupt junctions, a linear frequency variation with


voltage is obtained.
In the next subsections we will explore different topologies of varactor-tuned oscillators.
The choice of any one of them for a particular application will be made, considering the
specifications in terms of tuning bandwidth, frequency-tuning linearity, phase-noise or cost,
for instance.

7.5.2 Some Topologies of Varactor-Tuned Oscillators


The different varactor-tuned oscillator schemes, which will be presented here, are based on
the fixed-frequency oscillators already shown before in this chapter.

7.5.2.1 VCO based on the Colpitts topology


A voltage-controlled oscillator design can be made taking as a starting point the Colpitts
oscillator topology. The varactor diode can replace the capacitor C1 in the capacitive voltage
divider. Of course, additional changes must be made to the bias circuitry in order to guarantee
independent varactor and transistor bias (see Figure 7.21). By varying the varactor voltage,
typical tuning bandwidth of more than one octave can be achieved with some degradation of
the output power and phase noise within the tuning bandwidth.
Oscillators, Frequency Synthesisers and PLL Techniques 483

C3

C1
L
C2

Figure 7.21 Schematic of the voltage-controlled oscillator based on the Colpitts topology

C1 L

C2
C3

Figure 7.22 Schematic of the voltage-controlled oscillator based on the Clapp topology

7.5.2.2 VCO based on the Clapp topology


Because of the larger Q-factor achievable, the voltage-controlled oscillator based on the Clapp
topology will have lower tuning bandwidth but better phase noise performance and more
uniform output power over the bandwidth, compared to the VCOs based on the Colpitts
configuration. The frequency in a Clapp oscillator is normally tuned by varying the capacitor
in the series L-C resonator (see Figure 7.17). A practical configuration of a Clapp VCO
using a bipolar transistor and a varactor is given in Figure 7.22. In this example, the load has
been connected directly to the collector terminal.

7.5.2.3 Examples of practical topologies of microwave VCOs


The VCO explored up to this point uses well-known oscillator topologies, such as the Colpitts
or Clapp oscillators. Now, we will focus our attention on other practical oscillator topologies,
conceived specifically to achieve good frequency tuning performance using varactor diodes
as tuning elements.
Let us consider the general topology of a microwave oscillator considered above
(Figure 7.11). Simulations performed under small-signal conditions can serve to compare
the tuning bandwidth obtained by inserting the varactor diode into different branches of this
generic oscillator. The more common solutions place the varactor connected in the gate or
source terminals when using a FET as the active device, or to the base or emitter if a bipolar
transistor is used. The schematics of these VCOs are given in Figure 7.23 and Figure 7.24,
respectively. In both cases, only series feedback as been considered (Z3 = ∞) and the load
has been connected to the drain (or collector) terminal.
484 Microwave Devices, Circuits and Subsystems for Communications Engineering

RL

Figure 7.23 VCO with the varactor connected to the transistor’s base

ZL

Figure 7.24 VCO with the varactor connected to the transistor’s emitter

Self-assessment Problems
7.4 Calculate the tuning bandwidth of a Colpitts-based voltage-controlled oscillator,
considering that Cmax/Cmin is 10 for the given varactor.

7.5 Calculate the tuning bandwidth of a Clapp-based voltage-controlled oscillator,


considering that Cmax/Cmin is 10 for the given varactor.

7.6 Oscillator Characterisation and Testing


An oscillator is a system transforming part of the energy supplied in the DC region into a
sinusoidal high-frequency signal. This signal is, ideally, a single spectral line with zero width
and finite energy. In fact, real oscillators produce spectral lines of finite width, having random
amplitude, frequency and phase fluctuations accompanied by harmonics and sub-harmonics
of the spectral main line. These harmonic signals are generated by the non-linearities in the
oscillator.
Summarising, the oscillator output presents deterministic signals, the carrier and its
harmonics, characterised by its frequency and power, and the random components which
modulate the carrier producing noise sidebands which, in their turn, are determined by phase
noise measurements.
Oscillators, Frequency Synthesisers and PLL Techniques 485

In this section, we will present a brief review of the principal characteristics determining
the performance of a microwave oscillator and the basis of the experimental techniques used
to measure them.

7.6.1 Frequency

The frequency of the oscillating signal, or carrier frequency, is normally determined by


comparison with a high-precision oscillator, such as is performed by frequency counters.
The precision sources can require the use of atomic-frequency standards, only available
in very specialised laboratories, to obtain greater accuracy in the determination of the
frequency.
In microwave frequency counters, the microwave carrier is translated to lower frequencies
in order to be compared with the high-precision source. The spectrum analyser provides
the complete spectrum of the oscillator output and can be employed to estimate the carrier
frequency. The accuracy can typically be of 1 Hz in the case of frequency counters or, when
using spectrum analysers, of the order of a few hundreds of Hz, for measurements in the
range of radio frequencies, or about 1 kHz, when working in the microwave bands.

7.6.2 Output Power

The output power of an oscillator is considered to be the power that the oscillator delivers
to the load at the fixed carrier frequency, avoiding the contributions of unwanted noise
sidebands and harmonics. A first estimate of the carrier power can be obtained from a
spectrum analyser. Nevertheless, these measurements do not offer, in general, amplitude
errors better than 2 dB at microwave frequencies or 1.5 at radio frequencies. The accurate
measurement of power is not a simple issue because it involves specific procedures and
equipment.
Among the different methods of detecting power, the thermistor, thermocouple and
detector diodes are those most commonly employed for measurements in normal applica-
tions. These sensors, accompanied with precision attenuators for the highest power levels,
cover a wide dynamic range, from −70 dBm to a few watts. At microwave frequencies,
absolute accuracy of a few tenths of dB are achievable using precision diode detectors.

7.6.3 Stability and Noise

The frequency instabilities in an oscillating signal can be related to long-term and short-term
fluctuations. The long-term stability problems, expressed in parts per million of frequency
change per unit time, are related to the ageing process in the materials employed and with
environmental changes such as in the ambient temperature, pressure, etc. The short-term
fluctuations are related to frequency deviations from the nominal frequency during periods
less than a few seconds. These instabilities can be modelled as amplitude or frequency
modulations produced by deterministic phenomena, such as the influence of external AC
electromagnetic signals or other vibrating signals, or by random fluctuations related to internal
or external noise sources.
486 Microwave Devices, Circuits and Subsystems for Communications Engineering

∆A: random amplitude


contribution
∆␪: random phase
∆A contribution
noise phasor
resultant phasor
␻m

∆␪
carrier phasor

resultant phase angle

␻o

Figure 7.25 Vector diagram showing amplitude and phase noise in an oscillator

7.6.3.1 AM and PM noise


The oscillator output can be considered to be the result of the addition of a random noise
component to the noiseless phasor representing the carrier (see Figure 7.25). When the noise
component is parallel to the carrier, the vector sum only alters the amplitude of the oscillating
signal, resulting in the amplitude modulation noise or AM noise. If the noise component
is perpendicular to the carrier, it produces phase noise (PM noise). In general, the oscillator
noise close to the carrier is mainly phase noise, with the amplitude noise level very low. This
is because the limiting mechanism in the oscillator reduces the amplitude variations imposed
on the carrier.
Phase noise is a very important feature of oscillator design and characterisation. The phase
noise appears as sidebands with a continuous spectrum in a frequency range around the
nominal oscillating frequency. In order to illustrate the harmful effect of the oscillator phase
noise in different applications, various examples are given. For instance, in transmitters or
receivers of phase-modulated data, the phase noise will degrade the data recovery, increas-
ing the bit-error rate. In multichannel communication receivers the oscillator phase-noise
sidebands will be transferred, radian per radian, to the channels translated to intermediate
frequencies, producing problems of channel spacing between adjacent channels. In Doppler
radars, which measure the shift in frequency between the reference and the returned signals,
the phase noise limits the resolution and sensitivity.
If we observe the oscillating signal in a spectrum analyser, the phase noise appears as
sidebands with a continuous spectrum around the carrier or fundamental frequency, with
a spectral density decreasing with frequency offset. In fact, the carrier sideband spectral
density observed in any spectrum analyser can be considered to be phase noise only if the
amplitude modulation noise is negligible and the phase fluctuations are worse than that of
the local oscillator of the spectrum analyser. In general, this is the case for a wide range of
solid-state oscillators and it is the reason why the spectrum analyser can serve to determine
the phase noise.
Oscillators, Frequency Synthesisers and PLL Techniques 487

The phase noise at a given offset frequency from the carrier is measured as the value of
the spectral density in a 1 Hz window. The usual units are dBc/Hz, representing the spectral
density, referred to the power level at the carrier frequency. Since, when measuring using
a spectrum analyser, the measured level will depend on the detector resolution bandwidth,
the measured value must be normalised to 1 Hz in order to obtain a consistent estimate
of the phase noise. In this way, the phase noise is given by:

N
L ( fm) = (7.26)
C

where N is the noise power given in a 1 Hz bandwidth, after corrections, at fm Hz from the
carrier and C is the carrier power. Making the assumption of weak phase fluctuations, from
small angle modulation theory, L( fm) can be related to the phase deviation and Sθ ( fm), the
power spectral density associated to the phase fluctuations, through:

θ 
2
1.4θ RMS 
2
S ( fm)
L ( fm) =  peak  =  = θ (7.27)
 2   2  2

By considering the oscillator as an amplifier with feedback (see Figure 7.10), Leeson [3]
studied the phase noise in the oscillator. The phase noise defined in a 1 Hz bandwidth at a
given frequency offset fm from the carrier, produces a phase deviation given by:

RMS(Vnoise )
∆θ peak = (7.28)
RMS(Vcarrier )

In the preceding expression, Vnoise represents the noise voltage and Vcarrier is the carrier voltage
magnitude. The values of such magnitudes can be related to the white noise in the amplifier
and carrier power, giving:

FkT
∆θ peak = (7.29)
Pcarrier

where F represents the amplifier noise figure, k the Boltzmann constant and T the absolute
temperature. The spectral density of phase noise at the input of the amplifier is:

FkT
Sθ IN ( fm) = ∆θ 2peak = (7.30)
Pcarrier

The phase noise at frequencies close to the carrier shows a 1/f spectral density that can be
modelled as a phase modulator connected to the input of the amplifier. The spectral density
at the input can be written:

FkT  fcarrier 
SθIN ( fm) = 1 +  (7.31)
Pcarrier  fm 
488 Microwave Devices, Circuits and Subsystems for Communications Engineering

Considering the bandpass characteristic describing the resonator response, the closed loop
response of the complete oscillator gives the phase spectral density:

 1  fcarrier  
2

Sθ OUT ( fm) = Sθ IN ( fm) 1 + 2    (7.32)


 f m  2QL  

The overall phase noise is given by the expression:

 1  fcarrier   FkT 
2
fcarrier 
L( fm) = 1 + 2    1 +  (7.33)
 f m  2Q L  2 P
 carrier  fm 

This expression shows the different regions of the oscillator spectrum associated with the
upconverted 1/f and white noise and the white noise floor. The important role of the quality
factor Q in the minimisation of the oscillator phase noise can also be observed.
The residual phase modulation produced by the phase noise is an interesting item in
systems using phase or frequency modulations. From the previous expression giving the
phase noise, the residual phase noise modulation can be calculated as follows:

fh

∆θ = 冮
2 L( fm)dfm

(7.34)

where fh and fl are the highest and lowest frequency respectively of the frequency band in
which the phase noise spectrum is considered.

7.6.4 Pulling and Pushing


The pulling of an oscillator gives a measure of the change of the oscillating frequency which
is associated with variations of the value of the load impedance connected to the oscillator
output. The pulling for a load is specified by the magnitude of the return loss at any angle.
Pulling can be predicted by simulating the oscillator or by measuring the oscillation fre-
quency while varying the impedance load. It is usual to analyse the frequency change for a
constant magnitude of the load and to modify the phase from 0 to 2π radians. A return loss
magnitude of 12 dB is commonly used to define the pulling. This can be achieved, in a 50 Ω
system, by rotating a 29.9 or 83.5 Ω load connected to a variable length transmission line
having 50 Ω characteristic impedance.
The supply voltages generally show drifts with time, but also with temperature and imped-
ance fluctuations. The pushing of an oscillator determines how the oscillation frequency
changes with deviations from the nominal value of the bias voltages. Pushing is determined
by observing the variations in the oscillation frequency under different bias voltage conditions.
The result is expressed in units of frequency variation per volt unit.
When the supply voltage is noisy or when it shows a periodic oscillation, modulation of
the oscillator carrier may occur, resulting in a noticeable increase of the phase noise level. It
is very important to achieve a good filtering of these undesired signals in the supply voltages.
Oscillators, Frequency Synthesisers and PLL Techniques 489

Self-assessment Problems
7.6 The measured frequency sensitivity to variations in the supply voltage applied to
a given oscillator is 100 KHz/volt. Calculate the resultant peak phase deviation
and the modulation level for the resultant sidebands considering a supply ripple of
1 mV peak at 100 Hz.

7.7 Examining the expression given the oscillator’s phase noise, explain the origin of
the different regions that can be observed in the oscillator spectrum, of the slopes
and the corner frequencies for low Q and high Q cases.

7.7 Microwave Phase Locked Oscillators


Microwave solid state oscillators have bounded frequency stability. It depends on the reson-
ant structure or resonant network used as the load network connected to the active device.
The achieved stability of free running microwave oscillators is worse than the stability of
crystal quartz oscillators, normally used at much lower frequencies (in the range of MHz).
To enhance the frequency stability of microwave oscillators, phase locked loop techniques
are commonly used. In this approach a highly stable reference oscillator controls a micro-
wave oscillator. In this situation the microwave oscillator is synchronised by the reference,
this is known also as a synthesised microwave oscillator. When a microwave oscillator is
controlled in such way improvements in long-term (such as temperature), and short-term
frequency stability (phase noise behaviour) are obtained. Additional performance can be
included in microwave phase locked oscillators such as the possibility of frequency selection
by digital control. This capacity is very useful when adjusting local oscillator frequencies in
channelised communication systems.

7.7.1 PLL Fundamentals


A phase locked loop (PLL) can obtain the synchronisation of a microwave oscillator by
an external reference oscillator. The basic scheme of a PLL system is shown in Figure 7.26,
easily identifiable as a system with feedback control. The microwave oscillator must be
a VCO. A sample of the output signal is compared with the reference signal in a phase
detector. The output signal from it, the error signal, is a low frequency signal, which, after
amplification and filtering, is applied to the VCO tuning control. The frequency of the output
signal ( fout) in the case of Figure 7.26 equals N times the reference frequency ( fo).
PLL systems are feedback control systems and their analysis can be made through their
frequency response and loop stability. In a general analysis, a simplified system, as is
shown in Figure 7.27, is enough. In this case the reference frequency is the same as the
VCO frequency. This situation is not the normal situation in a microwave oscillator where
the reference frequency is very low compared with the microwave frequency. The simple
system of Figure 7.27 is very useful when one wants to know the behaviour of a general
PLL system.
490 Microwave Devices, Circuits and Subsystems for Communications Engineering

Phase
Reference Detector VCO
Fout
×N

fo
Freq. Loop Loop
Multiplier Amp Filter

Figure 7.26 PLL system to stabilise a microwave VCO

Phase
Detector VCO

Reference Loop Filter

Figure 7.27 PLL simplified scheme

F(s) VCO
(Kd) (K v)
θi(t)
θo(t)
vd(t) vc(t)

Figure 7.28 Signals under locking condition in a PLL

Assuming that a lock is established, the VCO frequency is identical to the reference fre-
quency. The output signal and input signal are only different in phase. Using the notation
shown in Figure 7.28 and assuming a sinusoidal response of the phase detector, the detector
output signal is vd (t).
If the phase difference is low, a straight line can approximate the sine function. This
approximation is called the linear model of a PLL system:

vd (t) = Kd sin[θi (t) − θo(t)] ≅ Kd[θi(t) − θo(t)] (7.35)

where θi (t) is the reference signal phase, θo(t) is the output signal phase, and Kd is the
detector constant, measured in (Volt/rad).
The loop filter is characterised by its transfer function F(s), while the voltage controlled
oscillator (VCO) can be characterised by its tuning slope, assuming a linear dependence
of frequency versus voltage, Kv describes such a slope with dimensions of (rad/sec)/(Volt).
As the frequency, f, is a measure of the phase change rate, the angular frequency ω (rad/sec)
can be obtained as:
Oscillators, Frequency Synthesisers and PLL Techniques 491

dθ (t )
2π f = ω = (7.36)
dt

and according to VCO tuning by the control voltage vc(t), it is also:

dθ 0 (t )
ω = Kvvc(t) = (7.37)
dt

Transferring time-based equations to the Laplace domain, the transform pair properties can
be applied:

dx(t )
x(t) ⇔ X(s) ⇔ sX(s)
dt

and Equation (7.37) is now:

Vc (s)
sθo(s) = KvVc(s) ⇒ θ o ( s ) = Kv (7.38)
s

Using these equations a PLL system is represented by the block diagram shown in
Figure 7.29, where Laplace transforms of signals are used.
From the block diagram several relations between signals are obtained:

Vd (s) = Kd[θi(s) − θo(s)]


Vc(s) = F(s)Vd (s)

By a combination of these equations, a new expression for output signal phase versus input
signal phase is obtained:

G(s)
θo(s) = θi (s) (7.39)
1 + G(s)

Kv
G(s) = Kd F(s) (7.40)
s

In Equation (7.40) G(s) is called the open loop gain because it takes into account the three
basic cascaded elements of the loop: phase detector, loop filter and VCO before closing the
loop. The open loop condition is very difficult to implement experimentally and is rarely
done. The open loop scheme is still valid for the analysis, and being the open loop gain is a

θi (s) + Vd(s) Vc(s) Kv


+ Kd F(s)
s θ o(s)

Figure 7.29 PLL block diagram using Laplace transform domain


492 Microwave Devices, Circuits and Subsystems for Communications Engineering

θ i(s) + θ i(s) θ o(s)


+ G(s) H(s)
θ o(s)

(a) (b)

Figure 7.30 PLL schemes: (a) open loop gain G(s); (b) closed loop transfer function H(s)

very important parameter for stability analysis. From Equation (7.39) the closed loop transfer
function H(s) can be defined as:

θ o ( s) K d Kv F ( s )
H ( s) = = (7.41)
θ i ( s) s + K d Kv F ( s )
Two basic schemes are shown in Figure 7.30 to identify the open loop gain G(s) and the
transfer function H(s) of a PLL system.
A new phase parameter, phase error, can be obtained as the difference between input and
output signal phase:
θe(s) = θi (s) − θo(s) = θi(s)[1 − H(s)] (7.42)

associated with an error transfer function defined as:

θ e ( s) s
H ( s) = = 1 − H ( s) = (7.43)
θi ( s) s + K d Kv F ( s )
From Equations (7.42) and (7.43) it can be concluded that both output and error signal
phases are filtered versions of the input signal phase, and all the analysis tools available for
linear systems can be applied here.
Using the usual nomenclature of control systems, a PLL system can be classified according
to its transfer function order, that is the highest order of the s variable in the H(s) denominator.
Due to the integrator action of the VCO, the system order is always the filter order plus one.
An additional classification of a PLL includes its type, that is the number of origin poles
(integrators) in the open loop gain. In general:

Kn (1 + a1s + . . . + al sl )
G(s) = (7.44)
s n (1 + b1s + . . . + bp s p )
where n indicates the type and n + p indicates the system order. As an example, a PLL
containing a loop filter with

1 + τ1s
F( s) = (7.45)
1 + τ2s
is an order two and type one system. Likewise, a PLL with a loop filter given by:

1 + τ2s
F( s) = (7.46)
sτ1
is an order two and type two system. Every PLL system is at least of order one and type one.
Oscillators, Frequency Synthesisers and PLL Techniques 493

20 log |G(jω)|

0 dB ω1 ω2 ω

Gain margin

arg G(jω)

ω1 ω2 ω

Phase margin
–180°

Figure 7.31 Bode plots for an unconditionally stable PLL showing phase and gain margins

7.7.2 PLL Stability


The PLL is a feedback system. It must be analysed for stability in the same way that
a negative feedback oscillator is studied. A PLL is unstable if it can meet the oscillation
condition. From the open loop gain G(s) analysis, if there is a frequency ω (rad/sec) having
G( jω) = −1, the system is unstable. A good method to check stability of control systems
is the use of Bode plots, or amplitude and phase responses of the open loop gain. In
Figure 7.31 Bode plots of an unconditionally stable system are shown. At frequency ω 1
crossing unity gain (0 dB) the phase must be higher than −180° for stability.
Phase and gain margins are indicative of the system’s proximity to becoming unstable.
The behaviour of a PLL system is not fully described by its stability condition. Other
important aspects are related to the system response to transients and the filtering properties.
A PLL must be able to achieve the locking condition in the turn on operation or when there
is a change in the reference frequency. The most critical element in the system is the loop
filter because it represents the major control that the designer can exercise over the PLL
response. The filter must have a low pass response to include the DC component, and it
must attenuate high frequency components. A good loop filter must attenuate as much as
possible the reference frequency (see Figure 7.26), in order to avoid a phase modulation of
the VCO at this frequency. On the other hand, the loop filter must have a wide enough
bandwidth to allow the PLL to respond to high speed changes.

7.8 Subsystems for Microwave Phase Locked Oscillators (PLOs)


Once the basis of understanding the operation of a phase looked loop has been established,
it is time to study each one of the ideal building blocks (see Figure 7.32). We will see how
494 Microwave Devices, Circuits and Subsystems for Communications Engineering

fxtl fref
I/R fout
PD LF VCO

1/N 1/V

Figure 7.32 Block diagram of phase locked oscillator

the required transfer functions can be physically implemented, which are the degrees of
freedom of the designer and what the criteria are to choose between different alternatives.
Some examples of components extracted from commercial catalogues will help us to get
closer the real world of the designer.
The VCO is characterised by a constant Kvco (MHz/volt), the phase detector (PD) by a
constant Kd (volt/rad) and the dividers by their division ratio (R,N,V). The loop filter (LF) is
defined by its poles and zeros.

7.8.1 Phase Detectors


The phase detector compares the feedback signal with the reference signal and provides a
signal proportional to the phase difference between the two inputs. In other words, the phase
detector is the comparator whose output is the error signal in the feedback loop. Nowadays,
except for specific applications like low noise tracking filters or high frequency loops, a
digital phase-frequency detector provides a better performance and a wider phase difference
range than the classic mixer operating as a linear phase detector within a limited range of
phase differences. It makes the phase-frequency detector an acquisition-aiding element.
The operation of a multiplier as phase detector (Figure 7.33) is explained below.
Let us consider two inputs with the same frequency and different time varying phases:

V1(t) = A cos(ω 0 t + φ 1(t))


V2(t) = B cos(ω 0 t + φ 2(t))

Vout(t) Low Pass Vd(t)


V1(t)
Filter

V2(t)

Figure 7.33 Multiplier as a phase detector


Oscillators, Frequency Synthesisers and PLL Techniques 495

cos (∆Φ)

1 Linear
zone

π
π 3π ∆Φ Phase difference
2 2

Figure 7.34 Approximate linear region of cosine function

The output of the multiplier with a multiplying factor K (physically implemented with a
balanced mixer, for example) is:
1
Vout(t) = KV1(t)V2(t) = ABK[cos(2ω 0t + φ1(t) + φ 2 (t)) + cos(φ1(t) − φ2(t))] (7.47)
2
The first term of the sum (with the double frequency) is filtered. The second term is what
matters because it contains the phase difference. If we plot the shape of a cosine function,
two kinds of linear regions versus phase difference can be distinguished, around 90° and
270°. For example around 270° (Figure 7.34) we can substitute:
cos(φ1(t) − φ2(t)) ≈ φ1(t) − φ2(t)
The constant factor in Vout is defined as the constant of the detector.
1
Kd = ABK (7.48)
2
It seems logical that the output must be independent of the input amplitudes. It means that A
and B must be as large as possible to operate the device in saturation mode.

7.8.1.1 Exclusive-OR gate


An exclusive OR logic gate can operate as a sequential phase detector, providing at the
output an average value proportional to the phase difference between the inputs.
In Figure 7.35 the scheme of the exclusive-OR gate is shown. X1 and X2 are the input
signals whose phase difference is evaluated. Y is the output of the logic gate. See the truth
table (Table 7.1). The time-averaged value is proportional to the phase difference between
X1 and X2.

X1
Y
X2

Figure 7.35 Exclusive-OR gate


496 Microwave Devices, Circuits and Subsystems for Communications Engineering

Table 7.1 Truth table of exclusive-OR gate

X1 X2 Y

0 1 1
0 0 0
1 1 0
1 0 1

Output
A

π 3π
− 2π
2 2
π π Φ(X1) – Φ(X2) = ∆Φ
2

–A

Figure 7.36 Transfer curve of phase detector

The transfer curve of this phase detector is plotted in the Figure 7.36. A is the maximum
amplitude of the output signal.
The timing diagram for several phase differences (0, π /2 and π) is shown in Figure 7.37.
The slope around each linear zone fixes the phase detector constant:
2A
Kd = (7.49)
π

7.8.1.2 Phase-frequency detectors


These can operate as frequency discriminators for large initial errors and then as coherent phase
detectors, once the system is within the range of lock. These two modes of operation require
some memory. Therefore, the phase-frequency detector is usually a digital circuit containing
flip-flops (Figure 7.38). The inputs are the reference and the signal coming from the VCO,
divided or not, depending on application. The outputs are usually called Up and Down.
The response of a digital phase-frequency detector is shown in Figure 7.39. The averaged
output amplitude of each port (UP and DOWN) and the difference are plotted versus phase
difference of the inputs.
The output of the circuit consists of a train of pulses whose duty cycle is proportional
to the phase difference between the two inputs. If this phase difference is more than 2π,
the polarity of the output signal depends on the frequency relation between the inputs. The
pulses are integrated later in the loop filter, providing a voltage whose variation corrects the
frequency of the voltage controlled oscillator.
The phase-frequency detector outputs (UP and DOWN) are active with a low level. If the
reference signal has a frequency higher than the VCO signal or if it leads in phase, then the
Oscillators, Frequency Synthesisers and PLL Techniques 497

π
∆Φ =
2
X1

X2

Y Maximum
output
Duty cycle δ = 50%
∆Φ = 0

X1

X2

δ = 0%
Y

∆Φ = π

X1

X2

Y δ = 100%

Figure 7.37 Timing diagram of phase detector

output UP consists of low level pulses and the output Down stays high. On the other hand
if the reference is delayed or the VCO has a higher frequency, then UP stays high and
DOWN pulses low. An example of timing diagrams is shown in Figure 7.40.
Four cases are considered: equal frequencies and reference leading or lagging, reference
slightly larger and slightly smaller than VCO. In normal operation there is a small oscillation
between the Reference and VCO signal phases and the VCO is continuously correcting its
frequency.
In Figure 7.41 the transfer characteristics (output voltage versus phase difference at the
input) of four phase detectors are summarised. The phase detectors are: a four-quadrant
multiplier (mixer), an exclusive or gate, an edge triggered master–slave flip flop and a digital
phase frequency detector.
A special type of phase detector is the Sampled Phase Detector, based on the multiplica-
tion of the reference before comparing. It means that the comparison is performed at high
frequencies. We will deal with them in the section on multipliers.
498 Microwave Devices, Circuits and Subsystems for Communications Engineering

Reference

Up

Down

Signal from VCO

Figure 7.38 Digital phase-frequency detector

Up

∆Φ
Down

∆Φ
Up – Down

–2π
0 2π ∆Φ

Figure 7.39 Response of digital phase-frequency detector


Oscillators, Frequency Synthesisers and PLL Techniques 499

REFERENCE

VCO

UP
0
DOWN 1
(a) ωREFERENCE ≡ ωVCO REFERENCE
LEADS IN PHASE

REFERENCE

VCO

UP 0
1
DOWN
(b) ωREFERENCE ≡ ωVCO REFERENCE
LEADS IN PHASE

REFERENCE

VCO

UP
0
DOWN 1
(c) ωREFERENCE ≡ 1.1 ωVCO

REFERENCE

VCO

UP 0
1
DOWN
(d) ωREFERENCE ≡ 0.9 ωVCO

Figure 7.40 Timing diagrams of digital phase-frequency detector


1 2 3 4 5 6 7
500

Output signals ud as a function of Can be controlled with


PD Type Signals Schematic diagram PD sensitive on Operating made
phase error ϑe frequency error ω1–ω 2 low pass filter Type . . .
u1
4. Quadrant-Multiplier ud
1 ud given by
1 u1 phase error
ud ϑe alone Phase linear
u2 u2
–π π π ω1–ω 2
Linear 1 –π
2 2
all
u2 can also be a square wave

ud
1 u1
Sine or square wave ϑe Phase quasi digital
in u2 –π π π
saturation –π
2 2

ud unsymmetric ud
u1 u1
O ϑe
2 u2 Phase digital all
–π π π
u2 –π ω1–ω 2
2 2
Q
EXCLUSIVE OR = ud = Duty cycle of the signal Q

ud ud Preferred ??
Q = UP
u1 –u UP R1 R 2 c
u1 ϑe u1 u1
–2π –π π 2π ω1–ω 2 Phase and digital PD
3 u2 frequency
u2 u2
Q Q = DOWN ud = weighted average of the –u DOWN
outputs UP and DOWN
underlined
JK–Master–
Edge triggered Slave–FF UP : weight +1 DOWN : weight –1 Probably controlled with
Case 1: U1 leading low pass filter type 3 having
U1 a pole at ω = 6 (Integrator)
ud ud
U2 G1 G2
UP
UP SFF1
u1 ϑe
DOWN 1 R 0 Phase and
0 frequency digital
4 –2π –π π 2π ω1–ω 2
Case 2: U2 leading
U1 RFF0 DOWN
u2 S 1
U2 ud = weighted average of the
outputs UP and DOWN
1
UP 0 G2 G4
UP : weight +1 DOWN : weight –1
DOWN
Microwave Devices, Circuits and Subsystems for Communications Engineering

Figure 7.41 Transfer characteristics of four phase detectors


Oscillators, Frequency Synthesisers and PLL Techniques 501

7.8.2 Loop Filters


The loop filter is the component of the PLL where the designer has the widest margin, so we
can consider it as the key component. The function of the loop filter is to reject unwanted
spurious signals, which could modulate the oscillator. The correcting signal is formed by
several components. The DC and low frequency components are responsible for the correc-
tion of the frequency-phase. Spurious components at the frequency of comparison or the
reference frequency modulate the oscillator and create unwanted side bands. If we talk about
digital phase detectors, the function of the loop filter is to integrate the output pulses of the
phase detector to control the frequency of the oscillator and its deviations.
The bandwidth of the loop filter also defines the bandwidth of phase noise improvement at
the output of the PLO, compared with the oscillator alone (free running).
The speed of the system response is inversely proportional to the bandwidth of the loop,
so a trade-off is necessary to choose the final value of this parameter.
As an example of a common loop filter, which is recommended by most of the synthesizer
ICs’ manufacturers, we will study in detail the 3rd order, type 2. This means the transfer
function in open loop has two poles at the origin (one due to the VCO and the other to the
filter) and the total number of poles is three. Why do we study such a complex filter and not
a simpler one, for example, with one pole? The criteria to choose the most adequate filter are
based on what error is acceptable for a fixed input signal (signal applied at the reference
port). The most common test signals considered at the input are a step in phase, a frequency
step, and a sloped frequency (parabolic phase).
The system must control the frequency of the oscillator according to the reference. It
means the steady state error (output phase minus input phase with time towards infinity)
must be zero for a phase step input and a frequency step input and bounded for a sloped
frequency. If an error is computed for each of the inputs (see PLL fundamentals) we can
verify that the minimum requirements to achieve it are: type 2 and 2nd order as can be seen
in Figure 7.42. For practical reasons it is recommended to add an additional pole in the
open loop which makes the order 3.
Next, the transfer function and the time constants of the loop filter will be deduced for
a PLL with a division ratio of N, a VCO constant Kvco and a phase detector constant Kd
(see block diagram in Figure 7.43).
The VCO adds a pole at the origin. It means that the filter must have two poles, one of
them at the origin:

1 + sT2
F( s) = (7.50)
sT1(1 + sT3 )

With the circuit topology shown in Figure 7.44 this transfer function can be implemented.

Vi: Input voltage


V0: Output voltage
Vi = R1I1
V0 = −(Z1 + Z2)I1
502 Microwave Devices, Circuits and Subsystems for Communications Engineering

Null error

t
(a)

Null error

t
(b)

Constant error

t
(c)

Figure 7.42 PLL error responses

fref fout
PD LF VCO

1/N

Figure 7.43 PLL with frequency division ratio of N


Oscillators, Frequency Synthesisers and PLL Techniques 503

C1 R2

R1 C2

Figure 7.44 Circuit diagram of 2-pole filter

The transfer function is the relation between V0 /Vi.

Vo − ( Z1 + Z 2 )
= (7.51)
Vi R1
Using Laplace expressions for the impedances, the transfer function becomes:

sR2C2 + 1 + R2 sC1
Z1(s) + Z2 (s) = (7.52)
sC1(sR2C2 + 1)

1 1 + sR2C2 + sR2C1 
F( s) = − (7.53)
sR1C1  sR2C2 + 1 

Compared with the generic expression, time constants can be obtained:
T1 = R1C1
T2 = R2(C1 + C2)
T3 = R2C2

Self-assessment Problems
7.8 Verify that the same transfer function can be implemented with the topology
shown in Figure 7.45. Verify the following set of equations which relates the new
time constants with the old ones.
Ta = 2T3 Ta = RaCa
Tb = T 1/2 Tb = RaCb
Tc = T3 Tc = RbCb
504 Microwave Devices, Circuits and Subsystems for Communications Engineering

Cb Rb

Ra Ra

Ca
+

Figure 7.45 Circuit of filter for exercise

The design procedure starts with the specification of the desired phase margin (m) and the
bandwidth, like in any other servomechanism. The Laplace variable s is substituted by jω in
the transfer function to obtain the frequency ω 0, which yields the phase margin. It can be
verified that in a first approach valid only in this type of filter, this value can be the desired
bandwidth.
The procedure continues by calculating the phase of the transfer function. Then 180
degrees are subtracted. This difference must be the desired phase margin and must reach
a maximum at ω 0, when the module of the open loop gain is 1 (it is just the definition of
the phase margin of a feedback system). The phase difference is derived and equalised
to zero to find the maximum.
The open loop transfer function is:

Kd Kvco 1 + jω T2 
G( jω ) H ( jω ) = − 1 + jω T  (7.54)
Nω 2T1  3

The phase difference with 180 degrees is:


φ = Arctan(ωT2) − Arctan(ωT3) + π − π
The derivative must be equal to zero:

dφ T2 T3
= − =0 (7.55)
dω 1 + (ω oT2 )2 1 + (ω oT3 )2

The frequency for the maximum phase difference ω 0 is:

1
ω o (φmax ) = (7.56)
T2T3
A relation between T2 and T3 is obtained:

T2 − T3
Tan(φmax ) = (7.57)
2 T2T3
Oscillators, Frequency Synthesisers and PLL Techniques 505

To calculate T1 the condition of the open loop transfer function equal to 1 is applied:

| G( jω 0)H( jω 0) = 1 | (7.58)

Finally, if the desired margin is m, the expressions of the time constants are the following:

sec(m) − tan(m)
T3 =
ω0
1/ 2
K K 1 + (ω oT2 )2 
T1 = d vco (7.59)
Nω o2 1 + (ω oT3 )2 

1
T2 =
ω o2 T3

Summarising, once the desired phase margin and the loop bandwidth have been specified for
a given loop, with some known values of Kvco, Kd and N, we can approximate the frequency
with open loop gain of unity and ωo for the loop bandwidth (valid only for the 3rd order,
type 2) and then obtain the time constants. Finally, a realistic set of resistance and capacitance
values can be chosen for the implementation of the real filter (for example, typical values of
resistance can be in the range from 10 Ohm to 100 kOhm, and typical values of capacitance
can be between 1 pF and 100 nF).

7.8.3 Mixers and Harmonic Mixers


A mixer can be employed in a more complex PLL in its typical role of frequency conversion
(see Chapter 5). This allows us to down-convert a high frequency to a lower value, suitable
to be applied to a frequency divider, for example, or combine the outputs of two PLLs to
achieve a small frequency step with a high frequency oscillator. If two signals with the same
frequency, but a slight phase difference are applied as local oscillator and radio-frequency
signals, an output signal basically proportional to the phase difference is obtained.
Based on this principle a common mixer can be used as phase comparator between two
signals. These two signals can be:

1. Main oscillator and reference without division (this is the case when the reference is a
remote carrier which comes from a transmitter and we try to track in a receiver)
2. Main oscillator divided by some integer N and reference oscillator divided by some
integer R.
3. Main oscillator and a harmonic of the reference. The harmonic can be generated by the
non-linearity of the mixer. In this case we talk about a harmonic mixer.

A harmonic mixer scheme is shown in the Figure 7.46. The frequency of the local oscillator
is 1816 MHz. The frequency of the VCO is 14168 MHz and is mixed with the 8th harmonic
of 1816 MHz, generated in the same mixer. This yields an intermediate frequency of 360 MHz
(14168-8 X 1816) which is then divided by 45 (Nt = 45) yielding a frequency of comparison
equal to 8 MHz.
506 Microwave Devices, Circuits and Subsystems for Communications Engineering

fcomp
:R
VCO fo
fR

NT G fOL

Figure 7.46 Harmonic mixer

7.8.4 Frequency Multipliers and Dividers


The PLL is a feedback system where frequency and phase are the parameters of interest.
The mixer provides the addition or the difference of frequencies, but the multiplication
or division of the frequency by a certain number (integer) can be also of interest in some
loop architectures. We have seen already an example of multiplication in the harmonic
mixer.
If a high frequency oscillator (GHzs) has to be built, we have to jump the gap between
this frequency and the reference (usually MHz). There area several ways to do it with some
advantages and disadvantages. The most commonly used is perhaps the division of the
main oscillator frequency by an integer N, allowing the comparison of the divided frequency
with the reference or with some divided version of it. This means we need a device able
to provide an output at a frequency fin /N where fin is the frequency of the input signal. N can
vary from 2 to 20,000 as an example, so the division is performed in several stages. The first
dividers, called prescalers, operate at high frequencies and usually divide by some power of
2 (2, 4, 8, . . . , 64, . . . ). They are a complicated and expensive part of the PLO. The
following dividers in the chain are less expensive because they operate at lower frequencies
and are built with cheap and well-known technologies.
We will see some kinds of dividers and the mode of operation. A loop based on the
division of the main oscillator frequency is really a multiplier of the reference frequency
and this can degrade the phase noise performance in some particular cases. One possibility
to overcome this drawback is the multiplication of the reference so that the comparison is
performed at the frequency of the main oscillator.

7.8.4.1 Dual modulus divider


This is a very popular topology of the variable ratio divider, implemented in commercial
circuits. It consists of three dividers commanded by a control signal (MC).
The first of the three dividers is a prescaler, able to operate at higher frequencies than the
rest. The block diagram appears in Figure 7.47.
Some preliminary conditions are the following:

• programmable dividers are counters, which count from the number they are programmed
to zero;
• M must be higher than A;
• the maximum division ratio is N(N + 1).
Oscillators, Frequency Synthesisers and PLL Techniques 507

fin/(M.N+A)
1/M

1/A

fin
1/N / 1/N+1

Figure 7.47 Dual modulus divider

Operation:

1. Both counters start at the same time and the control signal (MC) is ‘1’ until the first
counter finishes. The output starts with the value ‘1’.
2. When MC is ‘1’ the prescaler divides by N + 1. After finishing the A counter MC goes to
‘0’ and the prescaler starts to divide by N.
3. When the first counter finishes its count, A(N + 1) pulses of the input signal have happened.
4. When the M counter stops counting, N(M − A) input pulses have happened.
5. At that moment the output goes from ‘1’ to ‘0’.
6. During that time A(N + 1) + N(M − A) input pulses have generated a single transition
from ‘1’ to ‘0’.
7. The input frequency has been divided by A + NM.

With this procedure different values of division ratios can be achieved.


To understand how an electronic circuit can divide the frequency of an input signal, we
will see briefly a D-type flip-flop with feedback operating as divide by two (Figure 7.48) which
is the same as a multiplier by two of the period. Let us suppose the flip-flop is triggered by
the positive edge. The output tends to follow the data. The initial state is D = NQ = 1 and
Q = 0. When the clock rises, Q follows D and changes to 1. NQ changes to 0 and the same

ck

Q
D Q
NQ
ck NQ
D

Figure 7.48 D-flip-flop configured as a divide-by-2


508 Microwave Devices, Circuits and Subsystems for Communications Engineering

OUT 2
Fo = 9 QHz
COUPLER

VCO OUT 1
COUPLER

LOOP
SPD
FILTER

ACQUI LOCK
SITION DETECT
XTAL
100 MHz

Figure 7.49 9 GHz phase-locked oscillator

for D. The clock has to go down and rise again to produce the next change of state. Q goes
to 0 (present value of D) and NQ goes to 1 and the same happens to D. Looking at the timing
diagram it can be concluded that the period of the signal is multiplied by two.

7.8.4.2 Multipliers

To generate a harmonic of a given frequency a non-linear device is needed.


Depending on the order of the harmonic required different non-linearities are used.
With low orders (2,3,4) a MESFET transistor can be adequate. For higher orders, varactor
diodes are used. For the highest orders (over 20), step recovery diodes are the most suitable
choice.
In Figure 7.49 a Phase Locked Oscillator operating at 9 GHz is shown. The reference
(100 MHz) is applied to a Sampled Phase Detector (SPD) which multiplies the reference
by 90 (90 × 100 MHz = 9000 MHz) and then compares at that frequency.

7.8.5 Synthesiser ICs

Some IC manufacturers offer in their catalogues a product called a Synthesiser. What does
it mean? If I have to design a PLO operating from 2 to 3 GHz, for example, do I just have
to buy the adequate ‘synthesiser’, plug it in and that is all? I am afraid it is not so easy. The
chip called a synthesiser contains only some of the components of the PLO, for example,
a frequency divider for the reference, a prescaler and programmable dividers for the main
oscillator, the phase detector and some parts of the loop filter (i.e. the operational amplifier).
Some ICs have the option of using an external reference or generating the reference them-
selves by just adding a crystal resonator. The designer must add the voltage control oscillator
and the complete loop filter. Some procedure must be used to control the division ratio and
the output frequency, for example by parallel or serial programming.
Oscillators, Frequency Synthesisers and PLL Techniques 509

Vo

ωo f

Figure 7.50 Ideal output response of an oscillator

REF LEVEL
20.0 dBm

754.9910 MHz SPAN 10.0 MHz


VBW 300 Hz

Figure 7.51 Output spectrum of real oscillator with phase noise

7.9 Phase Noise


If we compute the Fourier transform of a sinusoidal function in the time domain we obtain
a delta function placed at the frequency of the sinusoid. See Figure 7.50.
Unfortunately it exists only in the ideal world of mathematics. If the output of an oscillator
is connected to a spectrum analyser a shape not as narrow as a delta will be observed (see
power spectrum in Figure 7.51). Phase noise is the cause of this.
First three definitions will be formulated:

Sφ ( fm): Spectral density of phase fluctuations, suitable to be used with the transfer function
in the linear analysis of a PLL.
L( fm): This is a measurement of the phase noise recognised by the US National Bureau of
Standards. Is the most intuitive way to express it.
σφ : Phase jitter is the standard deviation of the phase, considering the phase noise like
the result of a fixed oscillation with a random phase.
510 Microwave Devices, Circuits and Subsystems for Communications Engineering

Vo

ωo ωm f
ω o + 2π
ωm
ω o – 2π

Figure 7.52 Oscillator with sinusoidal modulation

Coming back to our ideal oscillator, if a sinusoidal phase modulation is applied to the oscillator
two side bands arise at a distance of the carrier fixed by the frequency of the modulating
sinusoid (see voltage spectrum in Figure 7.52). This provides us with a mathematical way to
understand and analyse phase noise.

V(t) = V0 sin(ω 0 t + θ (t))


θ (t) = ∆θ sin(ωm t)

The relative level of each side band is:

 ∆θ 
L( fm) = 20 log
 2 

The relation with the other parameters is given by:

1
L( fm) ≈ Sφ ( fm)
2

σ φ = ∫ Sφ ( fm )dfm

There are different sources of phase noise in an oscillator. Of course, if the oscillator is
phase locked, additional effects must be considered.

7.9.1 A simple model of phase noise: Leeson’s model


Let us consider an oscillator as an amplifier of noise with frequency selective feedback as
shown in Figure 7.53.
Considering the amplifier in base-band operation, a noise profile can be sketched (see
Figure 7.54).
This profile is upconverted with the feedback around the oscillation frequency (see
Figure 7.55).
We can suppose a certain noise figure of the amplifier F. By definition, the noise figure
is:
Oscillators, Frequency Synthesisers and PLL Techniques 511


F(ωm)

G
Nin + Nout

Figure 7.53 Model of oscillator as an amplifier with feedback

Slope = F−1
Sø(fm)

Flicker Corner

1E1 3E1 1E2 1E3 1E4 1E5 1E6


Baseband Offset Frequency

Figure 7.54 Baseband noise response

Nout
F= where G is the gain of the amplifier, Nin is the noise power at the input and
NinG
Nout is the noise power at the output.

If fo is the carrier frequency and fm is the offset frequency (far from the carrier) the peak
phase deviation would be:

FKT
∆φ p = with T absolute temperature and K Boltzmann constant.
Pavs

The RMS value would be:

1 FKT
∆φ RMS =
2 Pavs
512 Microwave Devices, Circuits and Subsystems for Communications Engineering

−20

SSB Phase Noise (dBc/Hz)


−60

F−3
−100

110 F−2
Flicker
Corner
F0 /2Q
180
1E1 3E1 1E2 1E3 1E4 1E5 1E6
Baseband Offset Frequency

Figure 7.55 Oscillator response with up-converted noise

If both sides of the carrier are taken into account f0 ± fm:

FKT
∆φ RMSTOT =
Pavs

The spectral density of phase fluctuation Sφ ( fm), would be:

FKT
Sφ (fm) = ∆φ RMSTOT
2
=
Pavs

If the bandwidth is B = 1 Hz, the product KTB will be equal to the noise floor of the amplifier
(corresponding to an offset frequency far from the carrier). With T room temperature:

KT = −174 dBm/Hz

For example, a feedback amplifier with a noise figure F = 6 dB and an available power
Pavs = 10 dBm will have a spectral density Sφ = −174 + 6 − 10 = −178 dBm/Hz.
If offset frequencies close to the carrier are considered, the shape of Sφ ( fm) shows a
1/f dependency empirically described by the corner frequency fc. It can be modelled like a
phase modulation at the input of the amplifier.

FKTB  fc 
Sφ ( fm ) = 1 +  (7.60)
2 PAVS  fm 
Oscillators, Frequency Synthesisers and PLL Techniques 513

The resonator is a band pass filter and it affects phase noise at the input. From the point
of view of the offset frequency, the resonator is a low pass filter with a transfer function
like this:

1
F(ω m ) = (7.61)
2Q Lω m
1+ j
ω0
Inside the bandwidth of the resonator the noise is transmitted without attenuation.
We can express this mathematically by:

 ω0 
∆φout ( fm ) = ∆φin ( fm ) 1 + (7.62)
 2Q Lω m 

Translating this expression to spectral phase density:

 f2 
Sφ out ( fm ) = 1 + 2 0 2  Sφin ( fm ) (7.63)
 f m(2Q L ) 

If we substitute for Sφ in( fm) we can obtain L( fm):

1 1 FKTB  f2  f 
L( fm ) = Sφ out ( fm ) =  1 + 2 0 2  1 + c  (7.64)
2 2 PAVS  f m(2Q L )   fm 

7.9.2 Free Running and PLO Noise


The phase locking not only allows the fine tuning of the oscillator. It also helps to reduce the
phase noise close to the carrier. How close to the carrier? It depends on the bandwidth of the
loop. Inside the bandwidth of the loop the phase noise is due to the reference multiplied by
the division factor. Outside the bandwidth the phase noise corresponds to the free running
oscillator. It means that the loop operates as a high pass filter for the oscillator phase noise
and as a low pass filter for the reference oscillator.
A certain bandwidth can be chosen to obtain optimum noise performance as is shown in
Figure 7.56, but this value may not be adequate for other requirements (such as reference
suppression, time of response of the loop).

L(f) VCO L(f) VCO L(f) VCO

REF REF REF

ωn < ωc ωn = ωc ωc < ωn

Figure 7.56 Optimum bandwidth for minimum noise is achieved with ω n = ω c


514 Microwave Devices, Circuits and Subsystems for Communications Engineering

7.9.1.1 Effect of multiplication in phase noise


If an oscillator operating at f0 is multiplied by N, the resulting oscillator at Nf0 has a phase
noise increased by N 2 :

L( fm)Nf 0 = N2L( fm)f 0 + A (7.65)

7.9.3 Measuring Phase Noise


The most intuitive way to measure the phase noise of an oscillator is to see the output
spectrum in a spectrum analyser. Nevertheless some understanding of the operation of the
spectrum analyser is needed.
The spectrum analyser operates as a receiver with a tuneable bandwidth filter. This band-
width sets the resolution of the measurement. It means that if we want to know the phase noise
10 kHz far from the carrier specified in dBc/Hz and the resolution bandwidth is 1000 Hz we
must add 10 · log(1000) = 30 dB to the value in dBs that is measured directly on the screen
between the carrier and the noise level 10 kHz from the carrier.
It is very important to have a spectrum analyser with a good local oscillator in terms of
phase noise to avoid errors in the measurement.

7.10 Examples of PLOs


An oscillator is synthesised to operate at a centre frequency 2.45 GHz with frequency steps of
250 kHz. A 2 MHz crystal is used as a reference. The VCO has a constant of 83.3 MHz/V.
The phase detector constant is 0.8 V/rad. The required phase margin is 45 degrees. A third-
order, type 2 filter is employed. The reference will be divided and a variable divider will
close the loop according to the general block diagram shown in Figure 7.32.
The loop is designed for a loop bandwidth of 50 kHz. The open loop gain and phase
response, the closed loop gain and error function and the time of response to a step of
2 MHz will be evaluated. Then the design will be repeated for a smaller bandwidth
(2.5 kHz) and the parameters will be compared.
Using the following phase noise data for the reference and the VCO (Table 7.2), the phase
noise performance of the complete loop can be compared for both bandwidths (50 kHz and
2.5 kHz).
First, we must establish the division ratio of the reference and the loop. If 250 kHz steps
are required the 2 MHz reference must be divided by 8. Therefore the comparison frequency
will be 250 kHz. It means that to fix the VCO frequency at 2.45 GHz the variable divider of
the loop must operate with a ratio of 9800.
A third-order type two loop is used, so the expressions for T1, T2 and T3 can be used. The
desired phase margin (m) is 45 degrees. Two different values are proposed for the loop
bandwidth (50 kHz and 2.5 kHz). Those values are below the comparison frequency because
the loop filter must reject the modulating side bands 250 kHz spaced from the carrier.
Both loops have two poles in the origin. For the 50 kHz bandwidth the zero is placed at
20.71 kHz (1/T1) and the pole is at 120.7 kHz (1/T3). With these values the desired phase
margin (45 degrees) is obtained. To achieve a smaller bandwidth (2.5 kHz) with the same
bandwidth the zero is moved to 1.035 kHz and the pole to 6.035 kHz.
Oscillators, Frequency Synthesisers and PLL Techniques 515

Table 7.2 Phase noise performance of reference and VCO oscillators

F offset L( foffset) (reference) dBc/Hz F offset L( foffset) (VCO) dBc/Hz

10 Hz −125 1 kHz −60


100 Hz −135 10 kHz −90
1 kHz −145 100 kHz −115
10 kHz −150 1 MHz −135
100 kHz −150 10 MHz −135

Table 7.3 Loop filter characteristics

ω 0 = 50 kHz ω 0 = 2.5 kHz

T1 (constant) 1.045e–6 s 4.1823e–4 s


T2 (zero) 4.8285e–6 s 9.6618e– 4 s
T3 (pole) 8.285e–6 s 1.6570e– 4 s

With these data the closed loop and open loop response are calculated for several fre-
quencies, verifying the desired phase margin and bandwidth. In Figure 7.57 the open loop
gain is plotted for the 50 kHz case. The phase response is plotted in Figure 7.58 showing the
desired phase margin in 50 kHz. Similar plots are obtained for the 2.5 kHz bandwidth.
The transient response to a 2 MHz step at the input can be obtained numerically show-
ing a faster response for the wider bandwidth filter (30 µs for the 50 kHz filter and 3.5 ms
for 2.5 kHz). In Figure 7.59 both responses are plotted showing a slower response for the
narrower loop. The response of the faster loop is plotted with a zoom of the time scale in
Figure 7.60. The phase noise profile of both loops is plotted in Figure 7.61. The effect of the
different bandwidth in the phase noise of the PLL is obvious.

Gain Margin undefined–system unconditionally stable

150
dB
100
dB
50
dB
100 1 10 100 1 10 100 1 10
0
dB Hz kHz kHz kHz MHz MHz MHz GHz GHz
−50
dB
−100
dB
−150
dB

Figure 7.57 Open loop gain with a bandwidth of 50 kHz


516 Microwave Devices, Circuits and Subsystems for Communications Engineering

Phase Margin = 45.0 deg. at 5.000E+04 Hz


100 1 10 100 1 10 100 1 10
0
Deg. Hz kHz kHz kHz MHz MHz MHz GHz GHz
−25
Deg.
−50
Deg.
−75
Deg.
−100
Deg.
−125
Deg. ωm
−150
Deg.
−175
Deg.
−200
Deg.

Figure 7.58 Open loop phase response with a bandwidth of 50 kHz

Fout (MHz)
3,000

2,500

2,000

1,500

1,000

500

0
0 500 1,000 1,500 2,000 2,500 3.000 3,500
t (us)

Fout (50 KHz) Fout (2.5 KHz)

Figure 7.59 Transient responses of two loop configurations


FoutFrequency
Oscillators, (MHz) Synthesisers and PLL Techniques 517
3,500

3,000

2,500

2,000

1,500

1,000

500

0
0 10 20 30 40 50 60 70 80 90 100
t (us)

Fout (2.5 KHz)

Figure 7.60 Transient response using 2.5 kHz loop showing more detail

L (foffset) dBc/Hz
0

−20

−40

−60

−80

−100

−120

−140
1 10 100 1,000 10,000 100,000 1,000,000 1.000E+07
foffset (Hz)

Lf (2.5 KHz) Lf (50 KHz)

Figure 7.61 Phase responses of two loop configurations


518 Microwave Devices, Circuits and Subsystems for Communications Engineering

References
[1] K. Kurokawa, ‘Microwave solid state circuits’, in Microwave Devices, John Wiley 8 Sons, Ltd, Chichester,
1978.
[2] K.M. Johnson, ‘Large signal GaAs MESFET oscillator design’, IEEE Transactions on Microwave Theory and
Techniques, Vol. 27, March 1979, pp. 217–227.
[3] D.B. Leeson, ‘A simple model of feedback oscillator noise spectrum’, Proceedings of the IEEE, February 1966,
pp. 329–330.
Index 519

Index

ABCD parameters 5, 164–6, 174, 208 available load power 216


acceptor band 16–17 available power gain 212
active balun 371 avalanche breakdown 22, 62
admittance 5, 191, 208 Avantek AT41485 transistor 269
AlGaAs/GaAs HBT 81
alumina substrates 134 balanced amplifier 239, 244–6
AM noise (amplitude noise) 486–8 advantages 246
Ampère’s law 114 basic configuration 244
amplifier 5–6 disadvantages 246
amplifier design 209–310 principle of operation 245
block diagram at radio frequencies 268 balun topology 372
broadband 239–46 band model for semiconductors 14–17
high frequencies components 256–67 band pass filter 152, 381–2, 414–18, 513
low-noise 246–56 normalisation 431–5
input and output stability circles 269 broadband 432–3
using CAD software 268–71 narrowband 433–5
practical circuit considerations 256–90 response 432
small signal 267–76 non-symmetric 435
strategies 237–9 band stop filter 381, 418–21
see also stability and under specific normalisation 435–9
components broadband 436–8
amplifier gain, definitions 209–23 narrowband 438–9
amplitude distortion 102, 105 Bessel functions 375
amplitude modulation 28–2, 102 Bessel low pass prototype filter, design tables
analogue-to-digital converter (ADC) 1 447
anti-parallel diode pair 364 Bessel prototype elements 404–5
approximate synthesis design 155 Bessel response 382, 390–1
arbitrary functions of single variable 112 biasing circuit 268
atomic-scale magnetic dipoles 115 active 274, 279
attenuation 46, 50, 380, 385–90, 392–3, 395–7, design 277–9
400–1, 403, 412–13, 416, 419, 429 for microwave bipolar transistors 272–6
approximation 110 implementation 279–90
travelling wave with 101 inductor 282
attenuation constant 98 passive 272–4, 277–9
attenuation function 389 single power supply 278
audio/video amplification and filtering 1 Unipolar power supply 278–9
automatic gain control (AGC) 1 bilinear mapping 221–2

Microwave Devices, Circuits and Subsystems for Communications Engineering Edited by I. A. Glover, S. R. Pennock
and P. R. Shepherd
© 2005 John Wiley & Sons, Ltd.
520 Index

bipolar power supply 277–8 channel conductance 351, 372


bipolar transistors 82, 319, 371 characteristic impedance 98–9, 130, 147, 150,
access resistance 321 152, 154, 158, 215
passive biasing circuits 273 Chebyshev filter, frequency translation 411, 414
Bode plots 493 Chebyshev low pass prototype filter, design
boundary condition 118–19 tables 441–6
branch-line directional couplers 163–5 Chebyshev low pass to band stop filter
breakdown voltage 22 frequency translation 420
broadband amplifier design 239–46 impedance transformation 421
broadband matching 191–4 Chebyshev prototype elements 400–4
Butterworth filter, impedance transformation 409 Chebyshev response 382, 386–90, 402
Butterworth low pass prototype filter, design chromatic dispersion 132
tables 440 circle mapping 221–2, 306
Butterworth low pass to band pass filter circle of constant normalised resistance 178–9
frequency transformation 417 Clapp oscillator 477
impedance transformation 417 VCO based on 483
transformation 416 C-L-C sections 145
Butterworth low pass to band stop filter, Colpitts oscillator 476–7
frequency transformation 420 VCO based on 482
Butterworth prototype filter 399 common base configuration 280–1
elements 395–400 common collector configuration 281
Butterworth response 382, 385–6 common emitter configuration 281
common source transistor 369
CAD 8, 222, 229, 255, 270–1, 290–306 compensated matching 239–41, 243
analysis 296–8 complete effective electrical length of stub 140
see also specific techniques complex propagation constant 98
layout-based design 302–3 compound semiconductors 10
low-noise amplifier design 268–71 computer aided design see CAD
modelling 293–6 conductance 4
schematic capture of circuits 302 frequency dependence of 108–9
transistor oscillators 479–80 conduction band 15
capacitance 4, 43, 147, 150 conductivity 42, 118
capacitance tolerance 261 conductor loss 137–8
capacitor continuous wave (CW) modulation 102
equivalent circuit 260 conversion matrix of non-linear capacitance,
impedance characteristic 260 diode mixer theory 335
imperfections 260 co-planner microwave probe 205
in amplifier design 259–63 coupled microstrip lines 148–63
internal parasitics 260–1 coupled microstrips, approximate synthesis 153
low loss interdigitated 263 coupling factor 152
parameters 261 covalent bonding in semiconductor crystal 10
quality factor (Q) 261 critical field value 22
temperature coefficient 261 cross-talk 148, 200
temperature range 261 Curtice model 74–5, 79
types 261–3 curvilinear squares 119
carbon composition resistor 258
carrier continuity equation 17–18 D-flip-flop 507
carrier wave 103 decoupling components 280–1
cascade circuits 306 depletion MESFET 68
ceramic capacitor 261–2 detector 38–9
Index 521

dielectric constant 13, 108, 116 distributed amplifier 239


dielectric loss 108–9, 137–8 distributed capacitance 264
dielectric materials, low loss 109 distributed circuit equations 125
dielectric resonator oscillators 479 distributed circuit theory 94–9, 113, 122, 125
differential charge element 11 distributed element matching 187
differential equations 96–7 distributed parameters 95–6, 105
diffusion current 12 donor band 15–16
diffusion current density 12 double stub matching 189–91
digital signal processing 379 double-gate FET mixers 349–57
digital-to-analogue conversion (DAC) 1 equivalent circuit for small signal analysis
diode capacitance as function of reverse bias 352
27 final design 355–6
diode equivalent circuit model 24 IF amplifier 354
diode function 320 double-gate MESFET 366
diode mixer theory 331–41 double-mix superheterodyne receiver 311
conversion gain 338–9 drain current 341, 351
conversion matrix of complete diode 337 drain-gate capacitance 349
conversion matrix of linear resistance 336–7 drain mixer 367
conversion matrix of mixer circuit 337–8 small-signal equivalent circuit 346–7
conversion matrix of non-linear capacitance drain-source current 353–4
335 drain-source voltage 353
conversion matrix of non-linear resistance/ drift current 12
conductance 333 dual modulus divider 506–8
harmonic balance simulation 339–41
input/output impedances 338–9 edge-coupling 148
large signal analysis 339–41 effective relative permittivity 150
linear analysis 332–9 electric field 11, 20, 114–15, 117, 119, 126,
linear and non-linear networks under LO 128–9
excitation 340 IMPATT diode 62
non-linear capacitance under LO excitation electric field strength, Gunn diode 53
335 electromagnetic analysis 298
non-linear resistance under LO excitation 333 electromagnetic coupling 208
relationship between IF (0) and RF (1) ports electromagnetic laws 123
338 differential form 116
single-diode mixer solution 333 electromagnetism, principles of 114–16
diode realisations 8 electromotive force (EMF) 114, 123
diode types 3 electron density 11
direct search 300 electron mobility 11
directional couplers 148 electron movement 11
discontinuity 138–9, 208 elliptic band stop filter
models 139–45 frequency response 428
dispersion 101–2, 104 including transmission zeros 428
and accommodation in design approaches elliptic filter tables 407
132–5 elliptic filter translation 423–8
microstrip 208 band pass 421, 426
dispersion expressions 158 band stop 421, 426–8
dispersionless line 101 high pass 425
displacement current 13, 114 low pass 423
displacement current density 13 low pass to band pass 427
dissipation factor 261 low pass to band stop 427
522 Index

elliptic low pass prototype filter, design tables filter synthesis techniques 383
448–59 filters 7, 379–459
elliptic networks 406 band pass see band pass filter
elliptic PLPF 407 band stop see band stop filter
elliptic prototype elements 405 classification 381
elliptic response 382, 390–3 construction using microstrip 145–8
emitter grounding 288 definition 381
energy absorption 108 description 381–2
energy band diagram 16 design process 384
GaAs 52 fundamentals 379–85
metal semiconductor junction 33 high pass 381, 430–1
P-N junction 20 implementation 383
Schottky junction 34–5 low pass see low pass filter
enhancement MESFET 68 normalisation 429–39
equivalent current generator 253 order number 382
equivalent π-network 146 simulation 384–5
error function formulation 300–2 type of signal passing through 382
even mode 149–60 types 381
exclusive-OR gate 495 final drift velocity 11
flow graph
Fano’s limits 241–3 effect of load on input reflection coefficient
feedback amplifier 512 217
feedback control systems 489 generator connected to load 215
feedback oscillator 472 modified 219–21
Fermi energy level 15, 32 techniques 306
FET two-port with generator and load 218
small-signal equivalent circuit 344 foreshortened open end discontinuities 139–41
types and applications 321–2 forward bias
FET mixers 341–9 P-N junction 23–4
double gate see double-gate FET mixers Schottky junction 35–6
double-balanced 359–60 forward travelling wave 112
gate-pumped 346 forward wave 99
large-signal equivalent circuit 343 Fourier analysis 101
resistive 348 Fourier coefficients 344
RF and LO both applied at gate 342 Fourier series 347–8, 372
single-balanced 358–9 frequency components 101, 104
single-ended 341–9 frequency dependence
single-gate 341–2 of conductance 108–9
large-signal and small-signal analysis of line parameters 105–8
343–6 frequency limitations 135–7
small-signal equivalent circuit 345 frequency multiplier 30–1, 51, 506
see also monolithic mixers design 360
field-effect transistor see FET frequency response 382
filling factor 130 frequency synthesiser 7
film chip resistor 259
filter ladder network 408 GaAs 10, 52, 319
filter network with arbitrary load resistance 408 biasing circuit with Zener diodes 279
filter realisation 7 energy band diagram 52
filter response 7, 382–3 metal semiconductor field effect transistors
mathematical 385–93 3
Index 523

monolithic microwave integrated circuit mean electron velocity vs. applied electric
(MMIC) technology 364–5 field 54
Schottky-Barrier diodes 313 operating modes 55–6
varactors 481 oscillators 59
GaAs FET, quiescent points 277 quenched dipole layer mode 56–7
GaAs FET MIC amplifier 139 self-oscillations 54–5
GaAs FET transistor design 277–9 series equivalent circuit 57–8
GaAs MESFET 3, 66–75, 344, 348, 366 three-layer structure 53
capacitance-voltage characteristics 70 transit-time dipole layer mode 56
current-voltage characteristics 68–9 Gunn oscillator
Curtice model 74–5 coaxial 469
equivalent circuit model 71 design technique 469
frequency dispersion 73
gate-source capacitance 70 harmonic balance technique 297, 355
large signal equivalent circuit 74 harmonic mixer 360–4
output conductance 72 balanced 362–4
small signal equivalent circuit 71–3 conversion loss 363
small signal transconductance 72 PLLs 505
transconductance 69 single-device 362
transconductance time delay 73 Hartley oscillator 477
GaAs substrate 135 HBT 3, 65, 80–8, 319, 366, 469
GaAs transistors, active biasing circuit access resistance 321
280 capacitance-voltage characteristics 84–6
gain circle 222, 231–7, 271 current balance expressions 87
gain definitions 231–7 current-voltage characteristics 84
gain dependence on frequency 236–7 DC simplified equivalent circuit under
gallium arsenide see GaAs forward bias 320
Gauss’s law 114 depletion capacitance 85
genetic algorithm search 300 diffusion capacitance 85–6
Gilbert cell 359 equivalent circuit model 82
gradient search 299 forward bias 82–3
grounding techniques junction capacitance 85
metallic stud through holes 289–90 large signal equivalent circuit 82, 87–8
plated through holes 288 resistance 83
raised ground plane 288–9 reverse bias 83
topside ground plane 288 small signal equivalent circuit 86–7
group delay 105 HEMT 3, 65, 75–80, 321–2, 366, 369, 469, 471
group velocity 102–5 capacitance-voltage characteristics 78
Gummel-Poon forward bias 319 common gate 369
Gunn devices 8 cross-section 76
Gunn diode 3, 52–9, 467–9 current-voltage characteristics 76–7
accumulation layer mode 56 Curtice model 79
applications 58 gUmu vs. vUgsu curves 77
current vs. time 55 large signal equivalent circuit 78–80
electric field strength 53 mixer 366
electric potential vs. position 54 output resistance 77
equivalent circuit 57 performance 76
limited-space-charge accumulation (LSA) physical structures 76
mode 57 small signal equivalent circuit 78
manufacture 52 topology 76
524 Index

transconductance 77, 79 input matching network 268


upconverter 358 input reflection coefficient 217
heterojunction bipolar transistor see HBT intercept distortion 105
high electron mobility transistor see HEMT intrinsic diode capacitance 28
high impedance line 282 intrinsic resistance 42
high pass filter 381 intrinsic semiconductor 10, 15
normalisation 430–1
high power amplification (HPA) 1 Kirchhoff’s laws 95, 274
hole concentration 12
hole continuity equation 18 Lange microstrip coupler 161
hole current 12 Laplace expressions for impedances 503
hole movement 11 Laplace transforms 491
homojunction bipolar transistor 319 Laplace variables 503
hybrid coupler 163–5 law of induction 114, 124–5
hybrid (h) parameters 5, 165, 208 least Pth error function formulation 301–2
least squares error function 300
identical polarity 150 line transformer 192
immittance chart 181, 183 linear frequency domain analysis 296
IMPATT diode 3, 8, 59–64, 463, 467 load power 215, 218
applied voltage, carrier density and current 61 local oscillator frequency 372
device equations 62–3 longitudinal section electric (LSE) mode 132
doping profiles 60 longitudinal section magnetic (LSM) mode 132
electric field 62 loop filter 501–5
equivalent circuit 63–4, 463 characteristics 515
principle of operation 60–1 loss 100–1
schematic structure 61 mechanisms 137–8
IMPATT oscillator 463–5 lossless approximation 111
impedance 5, 166, 174–5, 208, 210, 214–15 lossless line 101, 125
compensation 191 low impedance line 282
environment 224 low noise amplifier (LNA) 1
Laplace expressions for 503 low pass filter (LPF) 268, 381–4
impedance–admittance transformation 182 design 7, 148
impedance matching 5, 91–2, 176–95, 208, 238, frequency scaling 410
250 frequency transformation 412
Smith chart 182–91 impedance scaling 405–9
using quarter wavelength line transformer 194 impedance transformation 409, 412
using single section transformer 194–5 low pass to band pass transformation 414–18
imput impedance, effect of load 216–17 low pass to band stop transformation 418–21
imput reactance 146 low pass to high pass transformation 412–14
inductance 4, 124 lumped elements 367
inductor 263–7 lumped network topology 146
bias network 282 microstrip 145–9
biasing circuit 282 network design 285–6
effect of resistance 265 normalisation 429–30
equivalent circuit 264 normalised characteristics 384
impedance characteristic 265 resonant network transformation 421–2
parasitics 264 semi-lumped configuration 286
practical considerations 263–4 using distributed elements 286
quality factor (Q) 265–7 see also prototype low pass filter (PLPF)
ratio of reactance to series resistance 265 lumped capacitor 159–60
Index 525

lumped element matching 182–7 microstrip T-junction 142–5, 208, 294–5


lumped network topology 146 elementary equivalent circuit 143
lumped-elements low-pass filter 367 reference planes and impedences 143–4
microstrip vias 141–2
magnetic field 114, 120, 123–4, 128–9 microwave circuit integration, candidates for
magnetomotive force (MMF) 114–15 127
Mason’s rule 215, 217, 252, 306–9 microwave communications transceiver and
matrix representations 152 receiver 1
maximum conversion gain 366–7 Microwave Design System (MDS) program 355
maximum power transfer theorem 213–16 microwave filter 7
Maxwell’s equations 117–21, 123 microwave frequency amplifier 6
MDS (Microwave Design System) program 355 microwave integrated circuit see MIC
mean carrier lifetime 17 microwave mixers 311
MESFET 65–75, 321–2, 342, 348, 356, 371–2, microwave oscillator 7
471, 476 microwave signal amplitude 39
depletion 68 microwave transistor amplifier 267–8
double-gate 349, 366 mid-band design 153, 156
enhancement 68 minimax optimisation, error function 301
gain dependence on frequency 236–7 mitred bends 142
model 227 mixer measurements 356–7
see also GaAs MESFET conversion gain vs. IF frequency 356
metal film resistor, impedance characteristic 258 conversion gain vs. RF frequency 356
metal oxide semiconductor FET (MOSFET) 321 LO/RF isolation vs. LO frequency 357
metal semiconductor junction, energy band low level intermodulation 374
diagram 33 measured and simulated IP3 357
metallised film capacitor 262 mixers 6, 39, 311–77
MIC 94, 136 devices for 313–22
MIC-oriented transmission line structures 127 general properties 311–12
mica capacitor 262 harmonic see harmonic mixers
micromanipulator 206 ideal 312
microstrip 4–5, 126–48 input power vs. output power 329
approximate analysis or synthesis 130–1 intermodulation performance 348
concept 126 intermodulation power 327–9
coupler 152 intermodulation products 323–6
design aims 129–30 linear approximation 329–31
dispersion 208 microwave 311
filter construction using 145–8 monolithic see monolithic mixers
geometry 128 noise figures 311
low-pass filter (LPF) 145–9 non-linear analysis 322–31
open end fringing field 141 PLLs 505
open-ended 138 resistive 348
parameters 140–1 signal components 312
physical width calculation 130–2 time-varying resistance 348
ring resonator 134 see also diode mixer theory; FET mixers
separation 150 MMIC (Monolithic Microwave Integrated
structures 127 Circuit) technology 6, 65, 107, 364–5
substrates 132, 136 modulating frequency 104
systems using 126 modulating waveform 104
wavelength 156 modulation envelope 103–4
see also parallel-coupled microstrip modulation theory 102
526 Index

Monolithic Microwave Integrated Circuits non-linear convolution analysis 297


(MMICs) 6, 65, 107, 364–5 non-linear time domain transient analysis 297
monolithic mixers 364–75 non-linearity base-emitter equivalent diode
characteristics of monolithic medium 365 function 320
devices 366 Norton equivalent circuit 247
double-balanced FET mixers 370–4 Nyquist’s criterion 227
single-balanced FET mixers 368–70
single-device FET mixers 366 odd mode 149–60
MOSFET 321 Ohm’s law 12, 315
multipliers 508 one-dimensional wave equation 112
one-port device 200, 210, 307
N-type doping 13–14 one-port network 168–71, 174–5, 214
negative feedback 239, 243–4 open loop gain 515
negative output resistance 229 open loop phase response 516
negative resistance 227–8, 238 operating power gain 213
amplifiers 58–9 opposite polarity 150
network analysers 195–208 optimisation process 298–302
calibration approaches 206–7 optimisation search methods 299–300
calibration kits and principles of error oscillation
correction 198–202 principles of 226
cross-talk 200 stability 466–7
development 195–6 unintended 223
directivity 200 oscillation amplitude 463
error terms 201–2 oscillation condition 224–7, 237
load mismatch 200 oscillation frequency 463, 465
main circuit blocks 196 oscillation point 463
principle of operation 196–8 oscillator 7, 461–518
receiver 198 as amplifier with feedback 511
signal source 197 basic scheme 462
source mismatch 199–200 characterization 484–8
systematic errors 199–200 design 461
tracking 200 dielectric resonator 479
two-port test set 197–8 distributed elements 478–9
network methods 4–5, 91–2, 163–75 electrical circuit 462
non-equal complex source and load frequency 485
impedances 174–5 fundamentals 461–3
revision of z, y, h and ABCD matrices Gunn diode 59, 470
164–6 harmonic balance 480
network parameters 5, 208 high frequency 506
neutralisation 236 ideal output response 509
noise figure 248–50, 254, 270–1 IMPATT 463–5
mixers 311 impedance 462
noise figure circles 254–5 microwave phase locked 489–93
noise model for two-port device 252 subsystems 493–508
noise performance negative resistance diode 467–9
minimum 255–6 noise 485–8
two-port devices 248 output power 485
noise response, baseband 511 pulling and pushing 488
noise sources 249 Q factor 481–2
noise temperature 248–50, 252 response with up-converted noise 512
Index 527

simulation techniques 480 output spectrum of real oscillator with 509


solid state microwave 461–7 performance of reference and VCO oscillators
stability 485–8 515
testing 484–8 quality factor (Q) 488
transistor 469–80 phase responses of two loop configurations 517
varactor-tuned (VCOs) 481–3 phase shifting 29–30
voltage-controlled 481–3 PIN diode topologies 47–9
with sinusoidal modulation 510 phase velocity 100
see also specific types effect of line resistance 110
output matching network 268 phosphorus-indium (PIn) semiconductors 52
PIN diode 40–51
P-type doping 14–18 applications 45
parallel-coupled bandpass filter 148 charge density distribution 41
parallel-coupled microstrip 149 equivalent circuit 43–4
coupler compensation by means of lumped forward-biased 41–3, 51
capacitor 159–60 normalised admittance 46
coupler directivity 158–9 manufacturing 44
cross-sectional dimensions 150–1 phase shifting, topologies 47–9
determination of coupled region physical power limiting applications 50–1
length 156 reverse-biased 40–1
frequency response of coupled region normalised admittance 46
157–8 structure 44
special couplers 161–3 switching applications 45
parallel coupling 148–9 topologies 45–7
PCBs 127 thermal equilibrium 40
permeability 116, 134 variable attenuation 50
permittivity 116, 127, 132 planar electromagnetic simulation 298
of free space 114 planar transmission structure 127
phase constant 98, 103 PLL system 1, 8, 489–92
approximation 110 basic schemes 492
phase detector 494–7 block diagram 491
multiplier as 494 combined outputs 505
timing diagram 497 error responses 502
transfer characteristics 500 frequency division ratio of N 502
transfer curve 496 harmonic mixers 505
phase distortion 105 mixers 505
phase fluctuation, spectral density 512 signals under locking condition 490
phase-frequency detector 496–7 simplified scheme 490
digital 498–9 stabilising microwave VCO 490
timing diagrams 499 stability 493
phase-locked loop see PLL system P-N diode
phase-locked oscillator (PLO) 8, 508 applications at microwave frequencies 26
block diagram 494 electric model 24–5
examples 514–15 manufacturing 25
phase loop oscillator 360 P-N junction 18–31
phase noise 486–8, 509–14 charge field and potential variations across
measuring 514 19
multiplication 514 decreased potential barrier 23
optimum bandwidth for minimum noise energy band diagram 20
513 forward bias 23–4
528 Index

increased potential barrier 21 receiver


reverse bias 21–5 architecture 311
schematic diagram 25 development 311
schematic representation 18 minimum noise design 255–6
potential difference 118 mixer front end 256
power amplifier network analyser 198
layout representation 304 reference voltage controlled oscillator (RVCO)
schematic representation 303 197
power factor (PF) 261 reflection coefficient 5, 30, 176–7, 179–80, 182,
power gain 213 200, 210, 214, 217, 270, 380, 394
available 212 resistance (R) 4, 42
operating 213 resistor
power quantities 211 equivalent circuit 257
power transfer gain 380 frequency characteristic 257–8
propagation constant 109 in amplifier design 256–9
prototype low pass filter (PLPF) 383–4 stability 287
definitions 394 types 258
design 393–405 resonator circuits, transistor oscillators 473–4
pseudo-Fermi levels 22–3, 36 return loss 381
PSPICE reverse bias
circuit description 386 diode capacitance as function of 27
configuration 385 P-N junction 21–5
simulator 384 Schottky junction 34
reverse travelling wave 112
Q circles 192–4 reverse wave 99
quadratic detector 38 RF circuit design
quality factor (Q) CAD approach 291–3
capacitor 261 classical approach 290–1
inductor 265–7 range of acceptable performance 305
network 192 RF devices 2–3, 9–89
oscillator 481–2 RF/microwave subsystems 2
phase noise 488 ring coupler 163–5
transistor oscillator 473–4
varactor diode 26 saturation drain current 355
quarter wavelength shorted stub 282–5 scattering (s)-parameters 5, 163, 208, 211–14,
quasi-Newton optimisers 300 381
quasi-static parameters 128–32 advantages of 171
quasi-TEM 149 ATF-36163 FET 240
quasi-TEM lines 94, 108 with matching circuit 240
analysis of 122 conversion into y, h or ABCD parameters 174
quasi-TEM microstrip 157 conversion into z-parameters 171–4
quasi-TEM mode 126, 128–32, 136 definition 166–8
quasi-TEM structure 157 measurements 199, 202
quasi-TEM transmission lines 207 discrete naked devices 203–4
on-wafer devices 205–6
radiation loss 137–8 packaged devices 202–3
radio frequency chokes (RFC) 280 one- and two-port networks 168–71
implementation in bias network 282–6 Schottky-Barrier diode 313
random search 299 non-linear analysis 327
rat-race coupler 163–5 non-linear equivalent circuit 313–14
Index 529

Schottky capacitance equivalent circuit 316 single stub matching 187–9, 295
Schottky diode 32–9, 481 skin depth 105–6
applications 37 skin effect 106, 137
capacitance characteristic 318 Smith chart 5, 176–82, 208, 224
current source characteristic 314 active elements 182
electric model 36 compressed 182, 184, 270
experimental characterisation 317–19 development 176
linear equivalent circuit 316 impedance matching 182–91
at operating point 314–16 normalising 186
manufacturing 37 properties of 180–1
mixing applications 39 solid-state devices 2
non-linear charge characteristic 316 SOLT method 207
non-linear junction capacitance 314 source grounding techniques 288–9
source current equivalent circuit 315 source impedance of two-port device 251–4
spectral representation 315–16 space-time function 100
Schottky diode-based subharmonic mixer 362–3 spacer design 203
Schottky junction 32 spectral density 253
energy band diagram 34–5 spectral phase density 513
equivalent circuit model 36 stability 223–39
forward bias 35–6 conditional and unconditional 228–9, 238
reverse bias 34 low freqency 287
semiconductor crystal numerical tests 230
covalent bonding in 10 stability circles 222, 229–30, 233
N-doped 13 static TEM design parameters 128–9
P-doped 14 static TEM design synthesis 130
semiconductor diode 3, 9 statistical design of RF circuits 303–6
semiconductor energy band diagram 15 step-recovery diode 3, 51
semiconductor properties 10–18 Stokes’s theorem 116
semiconductor transistors 9 straight wire inductor 263
semiconductors 2–3 superheterodyne receiver 311
band model for 14–17 superregenerative receiver 311
charge movement 11 surface wave frequency 135–6
conductivity properties 12 surface waves 135–7
doped 13–14 synthesiser ICs 508
series resistive loading 235
series resonant circuit 421 Taylor expansion 39
short circuit 141, 282–5 Taylor series 38, 110, 323, 327
shunt admittance 138 Telegrapher’s Equation 111–12
shunt capacitance 144 TEM field theory method for ideal case 113–26
shunt connection of lumped element 185 TEM line 4, 94, 108, 207
shunt end susceptances 147 coordinate system for analysis 117
shunt metallised hole 142 ideal 110
side-fringing equivalent distance 136 static solution 117
SiGe HBTs 319 time-varying fields 117
signal to noise ratio (SNR) 249 time-varying solution 119–21
signal transmission 4–5, 91–2 TEM mode 4, 126
silicon semiconductors 61–2 features of 121–2
simultaneous conjugate match (SCM) 231–3 field theory 122
single frequency component 97 TEM propagation mode 4
single layer air core inductor 266–8 TEM time varying wave 125
530 Index

TEM wave 113–14, 116 mode classes 94–5


coaxial line 124 structural classification 92–4
physical picture 123 transverse electromagnetic line see TEM line
thermal equilibrium 18–21, 32–3 transverse electromagnetic mode see TEM
thermal noise generation 246–7 mode
equivalent circuit 247 transverse microstrip resonance 136
Thévenin equivalent circuit 247 transverse resonance 135–7
thick films 259 travelling wave with attenuation 101
thin films 259 triple stub matching network 192–3
three-dimensional (3-D) electromagnetic two-conductor transmission line 95–9, 114
simulation 298 two-point noise as four parameter system
time varying field 126 250–1
total reactance 214 two-pole filter 503
transconductance 341–2, 351, 360, 367 two-port device 200–1, 208, 211–13, 233,
GaAs MESFET 69 235–6
transducer gain 211–12, 237 noise model for 252
expression for 218–21 noise performance of 248
transfer curve of phase detector 496 source impedance of 251–4
transfer function 217, 501, 503 two-port network 165, 168–71, 174–5,
open loop 504–5 209–10
transient response definitions 379–81
two loop configurations 516 two-port single-element device 307–8
using 2.5kHz loop 517 two-port three-element cascade 308–9
transistor feedback oscillators 8 two-port two-element cascade 308
transistor mountings 202–7
transistor oscillator 8, 469–80 unilateral approximation 220
achievement of negative resistance 472 unilateral figure of merit (U) 236
CAD 479–80 Unipolar power source 278
common topologies 475
design fundamentals 471–4 valence band 16–17
generic topology 472 Van der Pol characteristic 464
practical topologies 478 varactor diode 8
Q factor 473–4 applications 28
resonator circuits 473–4 circuit for harmonic generation 31
transistors 3, 65–88 electrical model 26–8
empirical models 65–6 equivalent circuit 27–8
large signal models 66 phase shifters 29
model types 65–6 quality factor (Q) 26
modelling 65–6 vector analysis 116
small signal models 65–6 voltage amplification factor 367
with general lossy output loading 235 voltage gain 380
transmission coefficients 5 voltage-controlled oscillator (VCO) 8, 481–3
transmission lines practical topologies 483
ABCD parameters 5, 164–6, 174, 208 topologies of 482–3
as network 166 varactor connected to transistor’s base 484
classes 93 varactor connected to transistor’s emitter
general considerations 92–5 484
high frequency operation 109–13 voltage feedback
lumped parameter model of elemental length biasing circuit 273–4
96 with constant base current 274
Index 531

wave solutions 96–7 maximisation by centring tolerance window


weighted mean phase velocity 156 and element constrained region 305
Wilkinson divider 245 of RF circuit 304
wirewound resistor 258 yttrium iron garnets (YIGs) 8

y-parameters 5, 164, 191, 208, 296 z-parameters 5, 164, 208


yield conversion of s-parameters into 171–4
for given tolerance window 305 Zener diode, GaAs biasing circuit with 279

Index compiled by Geoffrey C. Jones

You might also like